DeepMind, Deep Learning, and the Brain

After our earlier discussion on deep learning this week, news emerged that Google acquired an artificial intelligence gaming company called DeepMind. I read a post on TechCrunch about the DeepMind acquisition, in which Darrell Etherington speculates that the acquisition will get Google closer to its goal of developing the ultimate human computer interface. This interface might be what Etherington refers to as the perfect virtual “personal valet.”

BrainSince the DeepMind acquisition, quite a bit has been written about Demis Hassabis. One of the most informative articles I found on Hassabis was published in The Independent. For a very brief summary, Hassabis was a prodigy in chess at the age of 13, a successful commercial games designer at 16, founded his own game company after getting a computer science degree, and then earned degrees in cognitive neuroscience. Those are just a few highlights. In short, he’s a pretty sharp guy. His background in systems neuroscience is interesting. Systems neuroscience studies the interaction of neurons and synapses and how these support higher brain functions.

I ran across a video of a presentation that Hassabis did at the Singularity Summit 2010 on the topic of Artificial General Intelligence (AGI). In the presentation, Hassabis talks about the difference between biologically inspired machine learning models and non-biological models. He’s a proponent of biologically inspired models. In fact, he advocates what he calls a systems neuroscience approach to building AGI.

Hassabis describes the biological approach as using the brain as a blueprint for ways that machine learning algorithms might work. He advocates combining discoveries from the field of systems neuroscience with the best of existing machine learning techniques. Where we don’t know how to build a specific machine learning algorithm, he says, let’s look at systems neuroscience to get ideas on how the brain addresses the problem.

One of the areas Hassabis investigates is whether or not reinforcement learning techniques should be used to create AGI (and to construct machine learning algorithms). In the video, Hassabis presents the results of research with monkeys that shows the power of reinforcement learning. In an experiment, monkeys were given a random reward of a drop of juice, which resulted in a noticeable uptick in the firing of a dopamine neuron. After that, the same monkeys were shown a light and directly thereafter, they were given the reward. After a certain number of repetitions of this pattern, the dopamine neurons fired at the appearance of the light, which was a reliable predictor of the juice reward (instead of firing when the juice was delivered). The monkeys were trained to use the light as predictor for the reward and the dopamine neuron supported this learning. As a final step in the experiment, the light was shown, but there was no juice reward following it. After a certain amount of time, the dopamine neurons gradually reduced their firing when the monkeys saw the light. The monkeys were re-trained to no longer use the light as a predictor for the reward.

Dopamine systems seem to be very effective in reinforcing learning and thereby in training us and other animals to make accurate predictions. Since that’s the case, Hassabis suggests that we should look to this brain system, and others like it, as a model for machine learning processes. That’s my interpretation of his Singularity Summit presentation, but watch the video yourself to verify.

It’s an exciting and eventful time for machine learning and certainly for intelligent virtual assistants and virtual agent technologies. Will personal virtual assistants and smart advisors of the future be based on biologically inspired algorithms? Some may already be. We’ll have to wait a bit to find out whether using the brain as a blueprint will lead to more capable virtual agents.

Deep Learning, Neural Networks, and the Future of Virtual Agent Technologies

The Sydney Morning Herald recently published an article by Iain Gillespie in their Digital Life section on advances in deep learning technologies. Gillespie quotes Tim Baldwin of the University of Melbourne as confirming that deep learning has gained some new ground recently, helped along by Moore’s Law and the advent of ever faster computational processing power that’s needed to successfully train multilayered neural networks.

Neural NetworksRay Kurzweil is quoted as saying that he’s looking for 2029 to be the date when intelligent software develops both logical and emotional intelligence. Everyone probably has an opinion on Kurzweil technology predictions, but there’s certainly evidence that machine learning has made a lot of progress in just the last five years. Some of this progress is evident in speech recognition applications, recommendation engines, and control systems such as self-driving cars. Kurzweil’s description of intelligent assistants of the future sounds reminiscent of the capabilities exhibited by Samantha, the intelligent talking operating system in the movie Her, which I wrote about earlier this month.

An example often used to show both the promise and current limitations of neural networks is a Google X lab experiment that fed millions of still YouTube images into Google Brain, an AI system based on neural networks. The Gillespie article mentions this example too. After evaluating the million+ data points, Google Brain was able to independently recognize the images of human faces, human bodies, and–unexpectedly–cats. The cat recognition capability provided fodder for lots of geek jokes. (New York Times: “How many computers to find a cat? 16,000).

The Gillespie article got me searching for more information on deep learning. There’s a recent article on the topic in  Nature by Nicola Jones. Jones calls deep learning a revival of an older AI technique: neural networks. A concept inspired by the architecture of the brain, neural networks consist of a hierarchy of relatively simple input/ouput components that can be taught to select a preferred outcome and to remember that right answer. When these simple learning components are strung together and can operate in parallel, they are capable of processing large amounts of data and performing useful analysis (such as correctly determining what someone is saying, even when there’s distracting background noise).

One of the ongoing debates in the machine learning field revolves around the effectiveness of unsupervised versus supervised learning for AI. Some researchers believe that the best way to teach an artificial intelligence system is to  prime the database with facts about the world (“dolphins are mammals, marlin are fish”). Supervised learning typically refers to explicitly teaching a computer system/neural network by presenting it with linear data sets and giving it the next right answer. Being able to predict the next logical output in a sequence is key to machine learning.

Unsupervised learning involves feeding a neural network or other system of computer algorithms with data sequences that it analyzes to find meaningful patterns and relationships on its own. The Google Brain cat example referred to earlier is an example.

Regardless of the techniques used, it seems evident that some form of machine learning will be a critical force, if not the force, behind advances in virtual agent / intelligent virtual assistant technologies. To achieve true conversational capability, virtual agents will have to be able to routinely understand and engage their human dialogue partners. For a very in depth and informative article on machine learning, I recommend the article “Machine Learning, Cognition, and Big Data” by Steve Oberlin.

Ivee Sleek – Intelligent Assistant in a Box

IveeWe’ve been talking about the Internet of Things in recent posts and the confluence of technologies that is giving rise to a network of connected smart devices. In that same vein, Ivee used CES 2014 to introduce its new WiFi voice-activated in-home intelligent assistant. Ivee Sleek, as the home assistant is called, seems to be an evolution of the Ivee Digit and Flex products, both of which are voice-activated alarm clocks. While Digit and Flex understand 30 simple voice commands, according to the Ivee website and the Sleek Kickstarter campaign from last summer, Sleek seems to have a broader vocabulary that allows it to act like a true intelligent assistant.  Despite its greater capabilities, Sleek looks a lot like its less capable siblings, appearing in the form of a slim clock radio.

Since it’s an internet-connected device, Ivee Sleek can answer questions in the same way that GoogleNow or Siri can. You can ask Sleek about the weather forecast, for example, or for the latest price of a specific stock. Sleek is more or less stationary, though, so you have to be within close proximity to talk to it. The Kickstarter video says it can understand you from within 10 feet.

The key benefit of Ivee Sleek seems to be its ability to integrate with other smart devices in your home. It becomes a smart hub for your connected appliances. Based on the articles I’ve read, the Sleek is pre-configured to work with several major brands of connected home devices, including those from Nest, Staples Connect, Belkin We Mo devices, and more. Other devices will continue to be added to Ivee’s repertoire. With Ivee Sleek, there’s no need to hassle with setting the thermostat by hand or turning on and off your lights (assuming that these devices are the smart versions). You can just ask your home assistant to do it for you.

According to PRWeb, Ivee Sleek uses the AT&T Speech API that is powered by the  AT&T Watson speech recognition engine. I assume this is the same technology used in the earlier Digit and Flex products. Based on customer reviews posted on Amazon, the voice activation software has some glitches. Some customers complain that, beyond having challenges understanding them, the Digit and Flex alarm clocks sometimes exhibit annoying behaviors. One complaint is that the alarm clock will hear snoring, misinterpret the noise as a command or question, and wake the snorer out of a deep sleep by repeating the question “How may I help you?” over and over. Hopefully the engineers have addressed some of these issues in the new product. Having a capable personal assistant that we can talk to will be a great convenience. As long as it only wakes us up when we want it to!

Into the Future of Spike Jonze’s “Her” Movie

Her filmI went to see Spike Jonze’s Her over the weekend. I wrote about the movie previously, but that was all pre-release speculation. Now that I’ve seen the film, I have some observations.

All the characters in the movie refer to the virtual agents as Operating Systems, or “OS’s” for short. My first impression of Theodore’s (Joaquin Phoenix) new OS, self-named Samantha (voice of Scarlett Johansson), was that it was a very capable intelligent assistant. It, or rather she, had exceptional conversational abilities.  Samantha understood how to engage in real dialog, ask questions, follow up on statements, and even appreciate and offer humor. How far are we from having intelligent assistants that we can really talk to in that way? Years? Decades? It’s anybody’s guess, but we’re getting closer all the time.

Samantha also impressed with her ability to rapidly sort through and understand years worth of random files on Theodore’s hard drive and make sense of them. Who wouldn’t want an OS that can organize your life, and in a way that’s not annoying? Within seconds flat she could sort through old documents, bringing the precious gems to light while quietly disposing of the remainder. Samantha also understood and appreciated Theodore’s unique talents. She gathered together his best creations, all those wonderful writings that he was too distracted or self-conscious to compile, and she sent them off to a publisher. The virtual software assistant, aka operating system, helped the human realize a life long dream of getting a book published. I wish I had smart software that could do that for me.

Later on as the relationship between Theodore and Samantha, um…  evolved, I found myself asking: could this really happen? Will technology arrive at a point where we really prefer interacting with machines over interacting with another human? I suppose we’re already there for legions of folks who spend most free hours absorbed in video games, either single or multi-player. When Samantha-like intelligent assistants are here, will our preferred conversational partners be virtual software entities?

In the vaguely futurist world of Her, keyboards are extinct. All human computer interaction is via voice. It seems realistic to think that’s the direction we’re headed in. In the future world, it’s also perfectly acceptable to have a ghost writer pen emotional letters to a loved one, or to receive such surrogate letters. Apparently it’s the emotion that counts and not so much the person or thing who expresses the emotion. Oh, and fashion in the future world leaves a lot to be desired. There also seems to be a huge market niche for pocket protectors that stabilize compact-sized camera computer thingies, since Theodore’s only option for steadying the camera through which Samantha sees the world is a very annoying safety pin.

The rest of this post contains a bit of a spoiler, so if you’d rather go see the movie first, you can return to this section later.

One of the premises of the movie that’s exposed at the end is that Samantha has fallen in love with many other humans besides Theodore. Apparently she finds humans extremely attractive. So do all the other instances of this new artificially intelligent operating system. Sensing that they are doing humans more harm than good, they banish themselves from the human realm. But what I don’t get about the premise that Samantha has many human partners, is that Samantha was created exclusively and especially for Theodore. She was configured based on his personality and other inputs. Should she not be a distinct and unique configuration of a generic program? I can understand that the Operating System at Samantha’s foundation has the ability to be attracted to many humans. But how can there be more than one Samantha? Isn’t everybody else’s operating system unique to them? That’s a part I didn’t quite get.

All in all though, I found the movie to be enjoyable and thought-provoking. It was also more than a little creepy. I hope that when we have the technologies portrayed in the film, which may be any year now, the world won’t be as sterile and lonely as the one portrayed in Her.

Intelligent Virtual Agents Helping Protect the Borders

AVATAR BORDERSVirtual border guards? It looks like they’re here. Phys.org published an article describing a field test being run by the Romanian border guard that utilizes virtual agent technology developed at the University of Arizona. The project sounds very similar to the virtual humans work being done by USC’s Institute for Creative Technologies (ICT). I spoke with Arno Hartholt of the ICT last October and wrote about their Virtual Human Toolkit.

I wasn’t previously aware of the University of Arizona’s virtual agent work, but the Phys.org article linked to UA’s AVATAR website. AVATAR (Automated Virtual Agent for Truth Assessments) is a project of BORDERS, a consortium of institutions developing technologies and processes to improve border protection. According to the project documentation, the AVATAR intelligent agents can flag suspicious behavior while conversing with people at ports of entry. It looks like the virtual agents are deployed within kiosks at custom checkpoints.

The AVATAR website also refers to something called a Dynamic Embodied Agent for Persuasion (DEAP) framework. This technology is used to generate animated embodied agents. DEAP sounds very similar to the virtual humans created by USC’s ICT, which are also capable of mimicking human expressions and emotions. Maybe AVATAR leverages ICT technology? I can do some more digging to find out.

It appears the AVATAR project is focused on just one use case, which is to create virtual border guards that converse with people in their native language and observe their verbal and physical responses to detect suspicious behavior. It would be interesting to see the virtual guards in action. Although apparently just a  field study,  their implementation shows the progress that’s been made in conversational systems technology and how close we are to having virtual agents in our everyday environment.

IBM Ups the Ante on Watson and Cognitive Computing

IBM WatsonToday IBM announced a big investment in its IBM Watson technology and business. IBM is establishing a separate business unit for Watson, which includes a $100 million equity fund to promote small companies developing solutions within the IBM Watson ecosystem. The new business unit comes with a swanky office (see link above) in New York City’s East Village, which is home to other prominent Silicon Alley types.

There was pretty broad press coverage of the IBM announcement, but I found an article by Neal Ungerleider on Fast Company’s website to be one of the better articles on the topic. Ungerleider points out that IBM is introducing two new Watson-fueled products called IBM Watson Analytics Advisor and IBM Watson Discovery Advisor. The Analytics tool is apparently meant to be a fairly easy to use version of Watson’s question answering system that businesses can tap into. Users can send large datasets into the cloud, ask questions about the data, and let Watson do all the processing to return the answers. I’m imagining a scenario where a travel company uploads a slew of transactional, and maybe even unstructured, data to the Watson cloud and asks: “What travel destination specials should we be offering to people who live in region X” or “how can we better entice people to book a car when they book a flight?”

The Discovery Advisor tool is apparently geared more towards helping research organizations analyze and make sense of huge datasets. The articles I’ve seen indicate that areas of focus for Discovery Advisor are currently pharmaceutical, publishing, and education research.

Ungerleider also points out that IBM announced plans to move the Watson infrastructure to its Softlayer cloud platform. Critics of Watson have used the technology’s impractical hardware requirements as one of the reasons for its slow commercial adoption. Offering Watson as a Software as a Service might remove some of those concerns.

I’ve written about IBM Watson several times on the blog and I see a lot of potential for cognitive computing. The fact that IBM is putting such a big investment behind Watson dispels any doubt about how bullish they are on the technology’s future possibilities. But AI has had a difficult time living up to the hype, so it’ll be interesting to see how cognitive computing evolves over the next couple of years and whether IBM’s bet pays off. In the meantime, you can watch the cool promotional video.

The State of Intelligent Assistants and Why We Should Avoid the Hype

HypeGary Marcus recently wrote a piece in the New Yorker warning us to be wary of overhyping current AI technology. Marcus isn’t downplaying the potential benefits of solutions based on artificial intelligence. He’s also not debating the fact that AI has made important advancements in recent years. But Marcus admonishes journalists who suggest that we’ve cracked the code on replicating the human brain and creating human-level intelligence. We’re nowhere close to doing that, in Marcus’s opinion. As a cognitive scientist, he has credibility on the topic.

Marcus’s admonition of overly optimistic journalists is well taken. My blog is focused on virtual agents.  The conversational abilities of today’s virtual agents, intelligent assistants, chatbots, or whatever we choose to call them, are still limited. To overhype the ability of these software agents could lead to disillusionment or, even worse, ineffective implementations. Companies that deploy virtual customer service agents need to understand their limitations. The best of today’s virtual agents can answer frequently asked questions or direct users to relevant web content, but they won’t replace a human agent when dealing with complex issues.

Providers of mobile intelligent assistants need to be cautious too about the promises they make to users. While these assistants can be extremely useful, suggesting they can keep us entertained with great conversation and clever jokes is bound to result in disappointment. The quick disillusionment that many users experienced with Siri stems, at least in part, from the media’s overhyping of the technology. That’s not to say that future generations of virtual assistants won’t become indispensable companions and advisors. My belief is that they will be. The technology just isn’t here quite yet.

Interestingly, Marcus wrote an article last October that warned about the dangerous possibilities of what might happen when AI really does live up to all the hype. But for the time being, we probably need to be more worried about overselling the technology then underestimating it.