The People Behind Britain’s AI Research & Development

The UK has a great heritage in AI, stemming back to pioneers such as Alan Turing, one of the undisputed fathers of the field. Britain has some of the best AI research groups in the world, including Cambridge, Imperial and University College London (UCL), and is a growing centre for tech entrepreneurship. But companies specialising in AI are few and far between, and those that do exist tend to be focused in one particular area.

Google’s acquisition of DeepMind has shone a light on this relatively nascent commercial sector, and Ben Medlock, co-founder of AI firm SwiftKey, believes that the UK is capable of building sustainable AI businesses to rival the giants of the West Coast.

Some experts have warned that artificial intelligence could lead to mass unemployment. Dr Stuart Armstrong, from the Future of Humanity Institute at the University of Oxford, said computers had the potential to take over people’s jobs at a faster rate than new roles could be created.

He cited logistics, administration and insurance underwriting as professions that were particularly vulnerable to the development of artificial intelligence. However, Anderson said AI is not all about “hacking the workforce to pieces”. Rather it is about making individuals more productive, and making sure that “processes get applied, stuff is accurate, errors are eliminated, and compliance is met”. Analyst firm Gartner predicts that ‘smart machines’ will have a widespread impact on businesses by 2020.

Here are some British AI companies making headlines in the field:

VocalIQ
Founded by Blaise Thomson, Martin Szummer and Steve Young, VocalIQ is a Cambridge-based startup, formed to exploit technology developed by the Spoken Dialogue Systems Group at the University of Cambridge, UK. With expertise and toolkits covering speech recognition, spoken language understanding, dialog design, and speech synthesis the company provides customised spoken language interfaces to any device, and for any application including smartphones, robots, cars, call-centres, and games. In September 2015 VocalIQ was acquired by Apple.

SwiftKey
SwiftKey uses artificial intelligence to make personalised mobile apps. It is best known for the SwiftKey keyboard, which learns from each individual user to accurately predict their next word and improve autocorrect. Its machine learning and natural language processing technology understands the context of language and how words fit together. SwiftKey products were embedded on more than 100 million devices last year, and the company has just launched an app for iPhones and iPads called SwiftKey Note. The company behind SwiftKey was founded in 2008 by Jon Reynolds and Dr Ben Medlock.

Bloom AI
Bloom develops consumer artificial intelligence software that attempts to befriend humans. Its lead developer, Jason Hadjioannou, showcased a companion AI app for iOS at 2015’s TechCrunch Disrupt in London. The app uses artificial intelligence to learn about and bond with the user, proactively striking up conversations and remembering the personal interests of the individual. Bloom’s companion app is interacted with via natural spoken language and presents one of the most realistic speech synthesis engines currently available. Bloom AI is based in England and was founded by Jason Hadjioannou.

Celaton
Celaton’s inSTREAM software applies artificial intelligence to labour-intensive clerical tasks and decision making. Every day, businesses receive mountains of information via email and paper. InSTREAM learns to recognise different types of information and process it accordingly. It never forgets, and handles huge volumes of information at high-speed. Like a real person, it asks questions when it is not sure what to do. Andrew Anderson is the founder and CEO of Celaton.

Lincor
Lincor provides hospital bedside computers to entertain patients and engage them with relevant information and advice. This virtual personal doctor will constantly analyse live personal health data to enable preventative medicine and tailored lifestyle advice. During a hospital visit, the data will be further analysed by hospital AI, giving doctors a more complete and detailed picture. Enda Murphy is the founder and CTO of Lincor Solutions.

Featurespace
Featurespace has developed and sells two software products based on its predictive analytics platform. One is for fraud detection and the other for marketing analytics. Its products use advanced proprietary algorithms to exploit the vast amounts of customer interaction data that many companies collect, delivering insights that can help to detect and prevent fraud and prevent customer churn. Featurespace’s team is led by CEO Martina King (former Managing Director of Aurasma and Yahoo! Europe) and CTO David Excell, who co-founded the company with Professor Bill Fitzgerald, alongside Matt Mills (Commercial Director) and Simon Rodgers (Director of Engineering).

Darktrace
Darktrace uses advanced mathematics to automatically detect abnormal behaviour in organisations in order to manage risks from cyber-attacks. Unlike software that reads log files or puts locks on the technology, Darktrace’s approach allows businesses to protect their information and intellectual property from state-sponsored, criminal groups or malicious employees that many believe are already inside the networks of every critical infrastructure company. Darktrace was founded in Cambridge, UK, in 2013 by mathematicians and machine learning specialists from the University of Cambridge, together with intelligence experts from MI5 and GCHQ.

Deep Neural Networks for Acoustic Modeling in Speech Recognition

Most current speech recognition systems use hidden Markov models (HMMs) to deal with the temporal variability of speech and Gaussian mixture models to determine how well each state of each HMM fits a frame or a short window of frames of coefficients that represents the acoustic input. An alternative way to evaluate the fit is to use a feed- forward neural network that takes several frames of coefficients as input and produces posterior probabilities over HMM states as output.

Deep neural networks with many hidden layers, that are trained using new methods have been shown to outperform Gaussian mixture models on a variety of speech recognition benchmarks, sometimes by a large margin. This paper provides an overview of this progress and represents the shared views of four research groups who have had recent successes in using deep neural networks for acoustic modeling in speech recognition.

INTRODUCTION

New machine learning algorithms can lead to significant advances in automatic speech recognition. The biggest single advance occured nearly four decades ago with the introduction of the Expectation-Maximization (EM) algorithm for training Hidden Markov Models (HMMs). With the EM algorithm, it became possible to develop speech recognition systems for real world tasks using the richness of Gaussian mixture models (GMM) to represent the relationship between HMM states and the acoustic input. In these systems the acoustic input is typically represented by concatenating Mel Frequency Cepstral Coefficients (MFCCs) or Perceptual Linear Predictive coefficients (PLPs) computed from the raw waveform, and their first- and second-order temporal differences. This non-adaptive but highly- engineered pre-processing of the waveform is designed to discard the large amount of information in waveforms that is considered to be irrelevant for discrimination and to express the remaining information in a form that facilitates discrimination with GMM-HMMs.

GMMs have a number of advantages that make them suitable for modeling the probability distributions over vectors of input features that are associated with each state of an HMM. With enough components, they can model probability distributions to any required level of accuracy and they are fairly easy to fit to data using the EM algorithm. A huge amount of research has gone into ways of constraining GMMs to increase their evaluation speed and to optimize the trade-off between their flexibility and the amount of training data available to avoid serious overfitting.

The recognition accuracy of a GMM-HMM system can be further improved if it is discriminatively fine-tuned after it has been generatively trained to maximize its probability of generating the observed data, especially if the discriminative objective function used for training is closely related to the error rate on phones, words or sentences[7]. The accuracy can also be improved by augmenting (or concatenating) the input features (e.g., MFCCs) with “tandem” or bottleneck features generated using neural networks. GMMs are so successful that it is difficult for any new method to outperform them for acoustic modeling.

Despite all their advantages, GMMs have a serious shortcoming – they are statistically inefficient for modeling data that lie on or near a non-linear manifold in the data space. For example, modeling the set of points that lie very close to the surface of a sphere only requires a few parameters using an appropriate model class, but it requires a very large number of diagonal Gaussians or a fairly large number of full-covariance Gaussians. Speech is produced by modulating a relatively small number of parameters of a dynamical system [10], [11] and this implies that its true underlying structure is much lower-dimensional than is immediately apparent in a window that contains hundreds of coefficients. We believe, therefore, that other types of model may work better than GMMs for acoustic modeling if they can more effectively exploit information embedded in a large window of frames.

Artificial neural networks trained by backpropagating error derivatives have the potential to learn much better models of data that lie on or near a non-linear manifold. In fact two decades ago, researchers achieved some success using artificial neural networks with a single layer of non-linear hidden units to predict HMM states from windows of acoustic coefficients. At that time, however, neither the hardware nor the learning algorithms were adequate for training neural networks with many hidden layers on large amounts of data and the performance benefits of using neural networks with a single hidden layer were not sufficiently large to seriously challenge GMMs. As a result, the main practical contribution of neural networks at that time was to provide extra features in tandem or bottleneck systems.

Over the last few years, advances in both machine learning algorithms and computer hardware have led to more efficient methods for training deep neural networks (DNNs) that contain many layers of non-linear hidden units and a very large output layer. The large output layer is required to accommodate the large number of HMM states that arise when each phone is modelled by a number of different “triphone” HMMs that take into account the phones on either side. Even when many of the states of these triphone HMMs are tied together, there can be thousands of tied states. Using the new learning methods, several different research groups have shown that DNNs can outperform GMMs at acoustic modeling for speech recognition on a variety of datasets including large datasets with large vocabularies.

This review paper aims to represent the shared views of research groups at the University of Toronto, Microsoft Research (MSR), Google and IBM Research, who have all had recent successes in using DNNs for acoustic modeling. The paper starts by describing the two-stage training procedure that is used for fitting the DNNs. In the first stage, layers of feature detectors are initialized, one layer at a time, by fitting a stack of generative models, each of which has one layer of latent variables. These generative models are trained without using any information about the HMM states that the acoustic model will need to discriminate. In the second stage, each generative model in the stack is used to initialize one layer of hidden units in a DNN and the whole network is then discriminatively fine-tuned to predict the target HMM states. These targets are obtained by using a baseline GMM-HMM system to produce a forced alignment.

In this paper we review exploratory experiments on the TIMIT database that were used to demonstrate the power of this two-stage training procedure for acoustic modeling. The DNNs that worked well on TIMIT were then applied to five different large vocabulary, continuous speech recognition tasks by three different research groups whose results we also summarize. The DNNs worked well on all of these tasks when compared with highly-tuned GMM-HMM systems and on some of the tasks they outperformed the state-of-the-art by a large margin. We also describe some other uses of DNNs for acoustic modeling and some variations on the training procedure.

View the full PDF publication here.

[Geoffrey Hinton, Li Deng, Dong Yu, Abdel-rahman Mohamed, Navdeep Jaitly, Andrew Senior, Vincent Vanhoucke, Patrick Nguyen, Tara Sainath, George Dahl, and Brian Kingsbury]

AI startups & companies in the landscape of Machine Intelligence

The following piece on AI startups & companies was created by Shivon Zilis in late 2014 and could be missing some information at the time of posting.

The original article can be found on Shivon’s website here and the full resolution version of the landscape image can be viewed here.

—————- Start —————-

I spent the last three months learning about every artificial intelligence, machine learning, or data related startup I could find — my current list has 2,529 of them to be exact. Yes, I should find better things to do with my evenings and weekends but until then…

Why do this?

A few years ago, investors and startups were chasing “big data” (I helped put together a landscape on that industry). Now we’re seeing a similar explosion of companies calling themselves artificial intelligence, machine learning, or somesuch — collectively I call these “machine intelligence” (I’ll get into the definitions in a second). Our fund, Bloomberg Beta, which is focused on the future of work, has been investing in these approaches. I created this landscape to start to put startups into context. I’m a thesis-oriented investor and it’s much easier to identify crowded areas and see white space once the landscape has some sort of taxonomy.

What is “machine intelligence” anyway?

I mean “machine intelligence” as a unifying term for what others call machine learning and artificial intelligence. (Some others have used the term before, without quite describing it or understanding how laden this field has been with debates over descriptions.)

I would have preferred to avoid a different label but when I tried either “artificial intelligence” or “machine learning” both proved to too narrow: when I called it “artificial intelligence” too many people were distracted by whether certain companies were “true AI,” and when I called it “machine learning,” many thought I wasn’t doing justice to the more “AI-esque” like the various flavors of deep learning. People have immediately grasped “machine intelligence” so here we are. ☺

Computers are learning to think, read, and write. They’re also picking up human sensory function, with the ability to see and hear (arguably to touch, taste, and smell, though those have been of a lesser focus). Machine intelligence technologies cut across a vast array of problem types (from classification and clustering to natural language processing and computer vision) and methods (from support vector machines to deep belief networks). All of these technologies are reflected on this landscape.

What this landscape doesn’t include, however important, is “big data” technologies. Some have used this term interchangeably with machine learning and artificial intelligence, but I want to focus on the intelligence methods rather than data, storage, and computation pieces of the puzzle for this landscape (though of course data technologies enable machine intelligence).

Which companies are on the landscape?

I considered thousands of companies, so while the chart is crowded it’s still a small subset of the overall ecosystem. “Admissions rates” to the chart were fairly in line with those of Yale or Harvard, and perhaps equally arbitrary. ☺

I tried to pick companies that used machine intelligence methods as a defining part of their technology. Many of these companies clearly belong in multiple areas but for the sake of simplicity I tried to keep companies in their primary area and categorized them by the language they use to describe themselves (instead of quibbling over whether a company used “NLP” accurately in its self-description).

If you want to get a sense for innovations at the heart of machine intelligence, focus on the core technologies layer. Some of these companies have APIs that power other applications, some sell their platforms directly into enterprise, some are at the stage of cryptic demos, and some are so stealthy that all we have is a few sentences to describe them.

The most exciting part for me was seeing how much is happening the the application space. These companies separated nicely into those that reinvent the enterprise, industries, and ourselves.

If I were looking to build a company right now, I’d use this landscape to help figure out what core and supporting technologies I could package into a novel industry application. Everyone likes solving the sexy problems but there are an incredible amount of ‘unsexy’ industry use cases that have massive market opportunities and powerful enabling technologies that are begging to be used for creative applications (e.g., Watson Developer Cloud, AlchemyAPI).

Reflections on the landscape:

We’ve seen a few great articles recently outlining why machine intelligence is experiencing a resurgence, documenting the enabling factors of this resurgence. (Kevin Kelly, for example chalks it up to cheap parallel computing, large datasets, and better algorithms.) I focused on understanding the ecosystem on a company-by-company level and drawing implications from that.

Yes, it’s true, machine intelligence is transforming the enterprise, industries and humans alike.

On a high level it’s easy to understand why machine intelligence is important, but it wasn’t until I laid out what many of these companies are actually doing that I started to grok how much it is already transforming everything around us. As Kevin Kelly more provocatively put it, “the business plans of the next 10,000 startups are easy to forecast: Take X and add AI”. In many cases you don’t even need the X — machine intelligence will certainly transform existing industries, but will also likely create entirely new ones.

Machine intelligence is enabling applications we already expect like automated assistants (Siri), adorable robots (Jibo), and identifying people in images (like the highly effective but unfortunately named DeepFace). However, it’s also doing the unexpected: protecting children from sex trafficking, reducing the chemical content in the lettuce we eat, helping us buy shoes online that fit our feet precisely, and destroying 80’s classic video games.

Many companies will be acquired.

I was surprised to find that over 10% of the eligible (non-public) companies on the slide have been acquired. It was in stark contrast to big data landscape we created, which had very few acquisitions at the time.No jaw will drop when I reveal that Google is the number one acquirer, though there were more than 15 different acquirers just for the companies on this chart. My guess is that by the end of 2015 almost another 10% will be acquired. For thoughts on which specific ones will get snapped up in the next year you’ll have to twist my arm…

Big companies have a disproportionate advantage, especially those that build consumer products.

The giants in search (Google, Baidu), social networks (Facebook, LinkedIn, Pinterest), content (Netflix, Yahoo!), mobile (Apple) and e-commerce (Amazon) are in an incredible position. They have massive datasets and constant consumer interactions that enable tight feedback loops for their algorithms (and these factors combine to create powerful network effects) — and they have the most to gain from the low hanging fruit that machine intelligence bears.

Best-in-class personalization and recommendation algorithms have enabled these companies’ success (it’s both impressive and disconcerting that Facebook recommends you add the person you had a crush on in college and Netflix tees up that perfect guilty pleasure sitcom). Now they are all competing in a new battlefield: the move to mobile. Winning mobile will require lots of machine intelligence: state of the art natural language interfaces (like Apple’s Siri), visual search (like Amazon’s “FireFly”), and dynamic question answering technology that tells you the answer instead of providing a menu of links (all of the search companies are wrestling with this).Large enterprise companies (IBM and Microsoft) have also made incredible strides in the field, though they don’t have the same human-facing requirements so are focusing their attention more on knowledge representation tasks on large industry datasets, like IBM Watson’s application to assist doctors with diagnoses.

The talent’s in the New (AI)vy League.

In the last 20 years, most of the best minds in machine intelligence (especially the ‘hardcore AI’ types) worked in academia. They developed new machine intelligence methods, but there were few real world applications that could drive business value.

Now that real world applications of more complex machine intelligence methods like deep belief nets and hierarchical neural networks are starting to solve real world problems, we’re seeing academic talent move to corporate settings. Facebook recruited NYU professors Yann LeCun and Rob Fergus to their AI Lab, Google hired University of Toronto’s Geoffrey Hinton, Baidu wooed Andrew Ng. It’s important to note that they all still give back significantly to the academic community (one of LeCun’s lab mandates is to work on core research to give back to the community, Hinton spends half of his time teaching, Ng has made machine intelligence more accessible through Coursera) but it is clear that a lot of the intellectual horsepower is moving away from academia.

For aspiring minds in the space, these corporate labs not only offer lucrative salaries and access to the “godfathers” of the industry, but, the most important ingredient: data. These labs offer talent access to datasets they could never get otherwise (the ImageNet dataset is fantastic, but can’t compare to what Facebook, Google, and Baidu have in house).

As a result, we’ll likely see corporations become the home of many of the most important innovations in machine intelligence and recruit many of the graduate students and postdocs that would have otherwise stayed in academia.

There will be a peace dividend.

Big companies have an inherent advantage and it’s likely that the ones who will win the machine intelligence race will be even more powerful than they are today. However, the good news for the rest of the world is that the core technology they develop will rapidly spill into other areas, both via departing talent and published research.

Similar to the big data revolution, which was sparked by the release of Google’s BigTable and BigQuery papers, we will see corporations release equally groundbreaking new technologies into the community. Those innovations will be adapted to new industries and use cases that the Googles of the world don’t have the DNA or desire to tackle.

Opportunities for entrepreneurs:

“My company does deep learning for X”

Few words will make you more popular in 2015. That is, if you can credibly say them.Deep learning is a particularly popular method in the machine intelligence field that has been getting a lot of attention. Google, Facebook, and Baidu have achieved excellent results with the method for vision and language based tasks and startups like Enlitic have shown promising results as well.

Yes, it will be an overused buzzword with excitement ahead of results and business models, but unlike the hundreds of companies that say they do “big data”, it’s much easier to cut to the chase in terms of verifying credibility here if you’re paying attention.The most exciting part about the deep learning method is that when applied with the appropriate levels of care and feeding, it can replace some of the intuition that comes from domain expertise with automatically-learned features. The hope is that, in many cases, it will allow us to fundamentally rethink what a best-in-class solution is.

As an investor who is curious about the quirkier applications of data and machine intelligence, I can’t wait to see what creative problems deep learning practitioners try to solve. I completely agree with Jeff Hawkins when he says a lot of the killer applications of these types of technologies will sneak up on us. I fully intend to keep an open mind.

“Acquihire as a business model”

People say that data scientists are unicorns in short supply. The talent crunch in machine intelligence will make it look like we had a glut of data scientists. In the data field, many people had industry experience over the past decade. Most hardcore machine intelligence work has only been in academia. We won’t be able to grow this talent overnight.

This shortage of talent is a boon for founders who actually understand machine intelligence. A lot of companies in the space will get seed funding because there are early signs that the acquihire price for a machine intelligence expert is north of 5x that of a normal technical acquihire (take, for example Deep Mind, where price per technical head was somewhere between $5–10M, if we choose to consider it in the acquihire category). I’ve had multiple friends ask me, only semi-jokingly, “Shivon, should I just round up all of my smartest friends in the AI world and call it a company?” To be honest, I’m not sure what to tell them. (At Bloomberg Beta, we’d rather back companies building for the long term, but that doesn’t mean this won’t be a lucrative strategy for many enterprising founders.)

A good demo is disproportionately valuable in machine intelligence

I remember watching Watson play Jeopardy. When it struggled at the beginning I felt really sad for it. When it started trouncing its competitors I remember cheering it on as if it were the Toronto Maple Leafs in the Stanley Cup finals (disclaimers: (1) I was an IBMer at the time so was biased towards my team (2) the Maple Leafs have not made the finals during my lifetime — yet — so that was purely a hypothetical).

Why do these awe-inspiring demos matter? The last wave of technology companies to IPO didn’t have demos that most of us would watch, so why should machine intelligence companies? The last wave of companies were very computer-like: database companies, enterprise applications, and the like. Sure, I’d like to see a 10x more performant database, but most people wouldn’t care. Machine intelligence wins and loses on demos because 1) the technology is very human, enough to inspire shock and awe, 2) business models tend to take a while to form, so they need more funding for longer period of time to get them there, 3) they are fantastic acquisition bait.Watson beat the world’s best humans at trivia, even if it thought Toronto was a US city. DeepMind blew people away by beating video games. Vicarious took on CAPTCHA. There are a few companies still in stealth that promise to impress beyond that, and I can’t wait to see if they get there.

Demo or not, I’d love to talk to anyone using machine intelligence to change the world. There’s no industry too unsexy, no problem too geeky. I’d love to be there to help so don’t be shy.I hope this landscape chart sparks a conversation. The goal to is make this a living document and I want to know if there are companies or categories missing. I welcome feedback and would like to put together a dynamic visualization where I can add more companies and dimensions to the data (methods used, data types, end users, investment to date, location, etc.) so that folks can interact with it to better explore the space.

Questions and comments: Please email me. Thank you to Andrew Paprocki, Aria Haghighi, Beau Cronin, Ben Lorica, Doug Fulop, David Andrzejewski, Eric Berlow, Eric Jonas, Gary Kazantsev, Gideon Mann, Greg Smithies, Heidi Skinner, Jack Clark, Jon Lehr, Kurt Keutzer, Lauren Barless, Pete Skomoroch, Pete Warden, Roger Magoulas, Sean Gourley, Stephen Purpura, Wes McKinney, Zach Bogue, the Quid team, and the Bloomberg Beta team for your ever-helpful perspectives!

Disclaimer: Bloomberg Beta is an investor in Adatao, Alation, Aviso, Context Relevant, Mavrx, Newsle, Orbital Insights, Pop Up Archive, and two others on the chart that are still undisclosed. We’re also investors in a few other machine intelligence companies that aren’t focusing on areas that were a fit for this landscape, so we left them off.

For the full resolution version of the landscape please click here.

—————- End —————-

Shivon Zilis is an Investor at Bloomberg Beta. She is currently based in San Francisco.

Her website is www.shivonzilis.com

How to Create a Mind by Ray Kurzweil

How does the brain recognise images? Could computers drive? How is it possible for man-made programmes to beat the worlds best chess players?

Google’s Director of Engineering, Ray Kurzweil delivers an interesting look on the subject and offers a fascinating discussion of how a computer can (or can’t) replicate the human mind.

Professor Stuart Russell – The Long-Term Future of (Artificial) Intelligence

This was published on YouTube on May 22, 2015 by username: CRASSH Cambridge

 

The Centre for the Study of Existential Risk is delighted to host Professor Stuart J. Russell (University of California, Berkeley) for a public lecture on Friday 15th May 2015.

The Singularity: A Philosophical Analysis

David Chalmers is a leading philosopher of mind, and the first to publish a major philosophy journal article on the singularity:

Chalmers, D. (2010). “The Singularity: A Philosophical Analysis.” Journal of Consciousness Studies 17:7-65.

Chalmers’ article is a “survey” article in that it doesn’t cover any arguments in depth, but quickly surveys a large number of positions and arguments in order to give the reader a “lay of the land.” Because of this, Chalmers’ paper is a remarkably broad and clear introduction to the singularity.

Singularitarian authors will also be pleased that they can now cite a peer-reviewed article by a leading philosopher of mind who takes the singularity seriously.

Below is a CliffsNotes of the paper for those who don’t have time to read all 58 pages of it.

The Singularity: Is It Likely?

Chalmers focuses on the “intelligence explosion” kind of singularity, and his first project is to formalise and defend I.J. Good’s 1965 argument. Defining AI as being “of human level intelligence,” AI+ as AI “of greater than human level” and AI++ as “AI of far greater than human level” (super intelligence), Chalmers updates Good’s argument to the following:

  1. There will be AI (before long, absent defeaters).
  2. If there is AI, there will be AI+ (soon after, absent defeaters).
  3. If there is AI+, there will be AI++ (soon after, absent defeaters).
  4. Therefore, there will be AI++ (before too long, absent defeaters).

By “defeaters,” Chalmers means global catastrophes like nuclear war or a major asteroid impact. One way to satisfy premise (1) is to achieve AI through brain emulation (Sandberg & Bostrom, 2008). Against this suggestion, Lucas (1961), Dreyfus (1972), and Penrose (1994) argue that human cognition is not the sort of thing that could be emulated. Chalmers (1995; 1996, chapter 9) has responded to these criticisms at length. Briefly, Chalmers notes that even if the brain is not a rule-following algorithmic symbol system, we can still emulate it if it is mechanical. (Some say the brain is not mechanical, but Chalmers dismisses this as being discordant with the evidence.)
Searle (1980) and Block (1981) argue instead that even if we can emulate the human brain, it doesn’t follow that the emulation is intelligent or has a mind. Chalmers says we can set these concerns aside by stipulating that when discussing the singularity, AI need only be measured in terms of behavior. The conclusion that there will be AI++ at least in this sense would still be massively important.

Another consideration in favor of premise (1) is that evolution produced human-level intelligence, so we should be able to build it, too. Perhaps we will even achieve human-level AI by evolving a population of dumber AIs through variation and selection in virtual worlds. We might also achieve human-level AI by direct programming or, more likely, systems of machine learning.

Premise (2) is plausible because AI will probably be produced by an extendible method, and so extending that method will yield AI+. Brain emulation might turn out not to be extendible, but the other methods are. Even if human-level AI is first created by a non-extendible method, this method itself would soon lead to an extendible method, and in turn enable AI+. AI+ could also be achieved by direct brain enhancement.

Premise (3) is the amplification argument from Good: an AI+ would be better than we are at designing intelligent machines, and could thus improve its own intelligence. Having done that, it would be even better at improving its intelligence. And so on, in a rapid explosion of intelligence.

In section 3 of his paper, Chalmers argues that there could be an intelligence explosion without there being such a thing as “general intelligence” that could be measured, but I won’t cover that here.

In section 4, Chalmers lists several possible obstacles to the singularity.

Constraining AI

Next, Chalmers considers how we might design an AI+ that helps to create a desirable future and not a horrifying one. If we achieve AI+ by extending the method of human brain emulation, the AI+ will at least begin with something like our values. Directly programming friendly values into an AI+ (Yudkowsky, 2004) might also be feasible, though an AI+ arrived at by evolutionary algorithms is worrying.

Most of this assumes that values are independent of intelligence, as Hume argued. But if Hume was wrong and Kant was right, then we will be less able to constrain the values of a superintelligent machine, but the more rational the machine is, the better values it will have.

Another way to constrain an AI is not internal but external. For example, we could lock it in a virtual world from which it could not escape, and in this way create a leakproof singularity. But there is a problem. For the AI to be of use to us, some information must leak out of the virtual world for us to observe it. But then, the singularity is not leakproof. And if the AI can communicate us, it could reverse-engineer human psychology from within its virtual world and persuade us to let it out of its box – into the internet, for example.

Our Place in a Post-Singularity World

Chalmers says there are four options for us in a post-singularity world: extinction, isolation, inferiority, and integration.

The first option is undesirable. The second option would keep us isolated from the AI, a kind of technological isolationism in which one world is blind to progress in the other. The third option may be infeasible because an AI++ would operate so much faster than us that inferiority is only a blink of time on the way to extinction.

For the fourth option to work, we would need to become superintelligent machines ourselves. One path to this mind bemind uploading, which comes in several varieties and has implications for our notions of consciousness and personal identity that Chalmers discusses but I will not. (Short story: Chalmers prefers gradual uploading, and considers it a form of survival.)

Conclusion

Chalmers concludes:

Will there be a singularity? I think that it is certainly not out of the question, and that the main obstacles are likely to be obstacles of motivation rather than obstacles of capacity.

How should we negotiate the singularity? Very carefully, by building appropriate values into machines, and by building the first AI and AI+ systems in virtual worlds.

How can we integrate into a post-singularity world? By gradual uploading followed by enhancement if we are still around then, and by reconstructive uploading followed by enhancement if we are not.

References

Block (1981). “Psychologism and behaviorism.” Philosophical Review 90:5-43.

Chalmers (1995). “Minds, machines, and mathematics.” Psyche 2:11-20.

Chalmers (1996). The Conscious Mind. Oxford University Press.

Dreyfus (1972). What Computers Can’t Do. Harper & Row.

Lucas (1961). “Minds, machines, and Godel.” Philosophy 36:112-27.

Penrose (1994). Shadows of the Mind. Oxford University Press.

Sandberg & Bostrom (2008). “Whole brain emulation: A roadmap.” Technical report 2008-3, Future for Humanity Institute, Oxford University.

Searle (1980). “Minds, brains, and programs.” Behavioral and Brain Sciences 3:417-57.

Yudkowsky (2004). “Coherent Extrapolated Volition.”

AI / Artificial Intelligence

Artificial intelligence (AI) is the intelligence exhibited by machines or software. It is also the name of the academic field of study which studies how to create computers and computer software that are capable of intelligent behaviour. Major AI researchers and textbooks define this field as “the study and design of intelligent agents”, in which an intelligent agent is a system that perceives its environment and takes actions that maximise its chances of success. John McCarthy, who coined the term in 1955, defines it as “the science and engineering of making intelligent machines”.

AI research is highly technical and specialised, and is deeply divided into subfields that often fail to communicate with each other. Some of the division is due to social and cultural factors: subfields have grown up around particular institutions and the work of individual researchers. AI research is also divided by several technical issues. Some subfields focus on the solution of specific problems. Others focus on one of several possible approaches or on the use of a particular tool or towards the accomplishment of particular applications.

The central problems (or goals) of AI research include reasoning, knowledge, planning, learning, natural language processing (communication), perception and the ability to move and manipulate objects. General intelligence is still among the field’s long-term goals. Currently popular approaches include statistical methods, computational intelligence and traditional symbolic AI. There are a large number of tools used in AI, including versions of search and mathematical optimisation, logic, methods based on probability and economics, and many others. The AI field is interdisciplinary, in which a number of sciences and professions converge, including computer science, mathematics, psychology, linguistics, philosophy and neuroscience, as well as other specialised fields such as artificial psychology.

The field was founded on the claim that a central property of humans, human intelligence—the sapience of Homo sapiens—”can be so precisely described that a machine can be made to simulate it.” This raises philosophical issues about the nature of the mind and the ethics of creating artificial beings endowed with human-like intelligence, issues which have been addressed by myth,fiction and philosophy since antiquity. Artificial intelligence has been the subject of tremendous optimism but has also suffered stunning setbacks. Today it has become an essential part of the technology industry, providing the heavy lifting for many of the most challenging problems in computer science.

[Snippet updated from wikipedia.org – 16th September 2015: https://en.wikipedia.org/wiki/Artificial_intelligence]