Share and discuss this blog

Tuesday, May 31, 2016

Is IBM trying to kill off AI research by misusing the word "cognitive?"

Welcome to the Cognitive Era, says IBM’s advertising. I have been trying to figure out what that could mean. If you look inside IBM’s site you find they are proud of Cognitive Health and Cognitive Cooking to take two examples of what the any claims they make. (I was wondering what Cognitive Elder Care. might be) I have trouble knowing what these terms mean because I know what the word cognitive means, and therefore I am finding what IBM is saying incomprehensible.

Let’s start with a brief history of the word cognitive. The field of Cognitive Psychology began in the late 60’s. Until that time, oddly enough, “how the mind works” was not a subject studied in psychology. A journal with that name started about then and I published an article in that journal in 1972 in that journal’s 3rd volume.

In 1977, I helped start the field of Cognitive Science in an attempt to join together people from disciplines other than psychology, all of whom cared about how the mind works. “Cognitive” meant: human thinking. When I started a company (1981) and called it Cognitive Systems. I was trying to say that the programs we built were modelled on human thinking. Around that time, John Searle visited my lab for a week, and wrote a somewhat nasty article featuring the Chinese Room problem that I assume was meant to be an attack on me. He was attacking what was referred to then as the strong AI hypothesis that said that if a computer could do smart things, then it was thinking. This was never my position, but Searle talked more to my students during that week then he did to me, so I guess he thought I believed in the Strong AI hypothesis. I do not. 

I think that the human mind does many things and I want to know how it does them and I want to build computer programs that operate in the same way. I am interested more in people than in machines but I think that if we copied people on a computer we could have some machines that behave intelligently. I don’t actually think the machines themselves would know what they were doing or actually would be intelligent. I used to my AI classes that I was a “fleshist.” If a person said something I would think that that person was thinking, but if a machine said the same thing, I wouldn’t think that. Others disagree with me on this, but I have never been an advocate of the strong AI hypothesis.

Why am I saying all this now? I am trying to understand what IBM could possibly mean when it uses the word cognitive and announces that we are now in the “cognitive era”. Do they think they Watson is actually thinking? I certainly hope not.

Do they think that Watson is imitating how people think in some way? I can’t believe that they think that either. No one has ever proposed that machines that can search millions of pages of text are smart. Matching key words, no matter how well you do it, is not even a human capability much less one that underlies the human ability to think. 

When AI started, they were some major people associated with it, whom, of course I knew well. Marvin Minsky as interested in people first, machines second. Allan Newell was interested in people first and machines second, Herb Simon wanted to copy chess grand masters,   rather than build chess playing machines that won by being fast at search. Even John McCarthy, with whom I never agreed about anything, was trying to copy how the mind worked. I once asked him “how can you believe that the mind happens to work using a logic system invented in the 19th century?” (McCarthy thought all knowledge representation could be done using Predicate Calculus.)

That phrase, knowledge representation, is the right thing to think about. It is the cornerstone of what AI was always all about. We need to represent knowledge in some way before we can effectively use it in a computer program. AI people have always worried about knowledge representation.

But this idea seems have disappeared in recent AI work and does not exist at all in Watson. Now AI people worry about how many pages of text they can search and how match key words and phrases. (Take a look at what IBM says that Watson does in natural language processing and you will only hear about phrase matching.)

Back to Cognitive Health. I am very interested in getting computers to be able to be helpful in health care. Do I think that they can be helpful by searching millions of pages of text? Probably. 

But there are real questions about what can be done to help people using AI. I, for one, have many questions I would like to ask about drugs and health issues, as I age, and I find that asking a doctor isn't always helpful because not all doctors the answers, and asking a computer is sometimes helpful if it can match what you asked to some text that it happens to have. As I write this I have a question about a drug I am taking that no text I can find can actually  answer. I have been able to find an expert at a major hospital to ask this question and he told me that his my father was taking it, so he certainly thought it was safe. But my questions was more subtle than that, in part because it is a new drug and often little is really known about new (and highly promoted) drugs.  I really have no one to ask

Would I like a computer to be able to answer these questions? Of course. That is what AI was supposed to be all about. We always wanted to get computers to be really helpful using everyday English backed by a great deal of knowledge of a given domain.  But if IBM keeps claiming it has solved Cognitive Health, I am wondering how many people who might want to think up about new ways to represent  knowledge about how the body works and how drugs work, might stop working on what they care about and simply assume that IBM owns the turf and that there is no reason to try and compete with them. IBM is not trying to solve the problem I care about, which is getting access to knowledge that is easily comprehensible about problems everyday people actually have. A lot of that knowledge isn’t in any computer in the first place or is in academic journals, so all the key word search in  the world really will not help the average person much.

As for Cognitive Cooking, one of my PhD students  in the 80’s wrote a program called CHEF that reasoned from prior cases in order to invent new recipes using on the ingredients you happened to have on hand.  I am sure CHEF was better than the program that IBM is selling because it was based on case-based reasoning and not on matching key words.

IBM really has to stop saying Cognitive about  everything it is trying to sell. It is hurting our future because it is very likely to serve as a deterrent to more research on knowledge representation, real natural language processing and case based reasoning. These are important problems. They have not been solved and IBM needs to stop asserting that they are by claiming Watson to be “cognitive" when it actually does no thinking at all.

Sunday, May 15, 2016

should we fear AI or just fear the people who write about AI?

Ashok Goel, a professor at Georgia Tech, made the news this week with the revelation that one of the TA’s that he used in his AI course was actually an ”AI.”  Now, I have no reason to believe that Goel was trying to do something wrongheaded. I think he was just playing around. But the media love AI stories these days and have yet again led the public on a very wrong headed journey about what AI is and what it can be.

Could a chatbot be an effective TA? It could certainly beat a rather disinterested TA. There are many real TA’s who are not really very good or very interested in what they are being made to do. But that is not an important question. The right question is whether we could make a real AI program that could perform this function as well as a very good person does? The news media want us to believe we are on our way to that. I have a different point of view. It is not that we can’t do it. It is that we aren’t actually trying to do that yet. To explain this, I need to provide some history.

The first of what are now called chatbots was called ELIZA and was written by Joseph Weizenbaum. This is well known. What I will now tell you is not so well known. Weizenbaum was a friend of mine. Ken Colby was a friend of mine as well. I know some things about all this that others may not know.

Colby and Weizenbaum were friends, at least they started out as friends. Colby was a psychiatrist who became interested in whether computers could function as therapists. I was hired by Colby (in 1968) to help him at the point where he was building PARRY. PARRY was supposed to be a paranoid patient. Colby was interested in showing that he could fool therapists who would not know if they were talking with a real person or a computer. PARRY supposedly lived in a locked ward of a hospital and it was Colby’s idea to build an effective model of paranoia and prove it was effective by fooling therapists. In essence, he was trying to pass the Turing test with PARRY.

He and Weizenbaum talked about these issues for years and at some point had a falling out. Weizenbaum wanted to show up Colby by building program that could fool people at MIT into thinking it was a Rogerian therapist. He chose Rogerian  therapists, in part, to make fun of Colby, but also because Rogerians don’t usually say that much in response to what their patients tell them.

An example of ELIZA that I found on the web is:

How do you do. Please state your problem.
>> I spend too much time with computers.
Why do you mention computers ?
>> You are one.
Does it please you to believe I am one ?
>> Yes.
You seem to be quite positive.
>> I suppose I am.
How long have you been ?

Weizenbaum found that people wound up pouring their hearts out to ELIZA and he was very upset by this. He wrote a book attacking AI, called Computer Power and Human Reason, to explain why he thought AI would never work. The irony is, of course, that Goel’s program did no more than what ELIZA did in the 60’s (possibly even less), but it is now worthy of articles in the Wall St Journal and the Washington Post. Key word analysis that enables responses previously written by people to be found and printed out, is not AI. Weizenbaum didn’t think he was building a Rogerian therapist (or doing AI) really. He was having some fun. Colby was trying to model a paranoid because he was interested in whether he could do it. He did not think he was building a real (AI) paranoid. And, I assume Goel does not think he is building a real AI TA. But the press thinks that, and the general public will soon think that, if it keeps publishing articles about things like this.

This technology is over 50 years old folks. Google uses key words, as does Facebook, as does every chatbot. There is nothing new going on. But we all laughed at ELIZA. Now this same stuff is being taken seriously.

What is the real problem? People do behave in any way that is anything remotely like what these “AI’s” do. If you tell me about a personal problem you have, I do not respond by finding a sentence in my memory that matches something you said and then saying that without knowing what it means. I think about your problem. I think about whether I have any reasonable advice to give you. Or, I ask you more questions in order to better advise you. None of this depends upon key word and canned sentences. When I do speak, I create a sentence that is very likely a sentence I have never uttered before. I am having new ideas and expressing them to you. You say your views back to me, and a conversation begins. What we are doing is exchanging thoughts, hypotheses and solutions. We are not doing key word matching.

It may be that you can make a computer that seems paranoid. Colby had a theory of paranoia which revolved around “flare” concepts like mafia, or gambling, or horses. (See his book Artificial Paranoia.) He was trying to understand both psychiatry and paranoia using an AI modeling perspective.

The artificial TA is not an attempt to understand TA’s, I assume. But, let’s think about the idea that we might actually like to build an AI TA. What would we have to do in order to build one? We would first want to see what good teachers do when presented with problem students are having. The Georgia Tech program apparently was focused on answering student questions about due dates or assignments. That probably is what TA’s actually do which makes the AI TA question a very uninteresting question. Of course, a TA can be simulated if the TA’s job is basically robotic in the first place.

But, what about creating a real AI mentor? How would we build such a thing? We would first need to study what kinds of help students seek. Then, we would have to understand how to conduct a conversation. This is not unlike the therapeutic conversation where we try to find out what the student’s actual problem was. What was the student failing to understand? When we try to help the student we would have to have a model of how effective our help was being. Does the student seem to understand something that he or she didn't get a minute ago?   A real mentor would be thinking about a better way to express his advice. More simply? More technically? A real mentor would be trying to understand if simply telling answers to the student made the best sense or whether a more Socratic dialogue made better sense. And a real TA (who cared) would be able to conduct that Socratic dialogue and improve over time. Any good AI TA would not be trying to fake a Rogerian dialogue but would be thinking how to figure out what the student was trying to learn and thinking about better ways to explain or to counsel the student.

Is this possible? Sure. We stopped working on this kind of  thing because of the AI winter than followed from the exaggerated claims being made about what expert systems could do in1984. 

We are in danger of AI disappearing again from overblown publicity about simplistic programs.

To put this all in better perspective, I want to examine a little of what Weizenbaum was writing in 1976:

He attacked me (but started off nicely anyhow):

Roger C. Schank, an exceptionally brilliant young representative of the modern school, bases his theory on the central idea that every natural language utterance is a manifestation, an encoding of an underlying conceptual structure. Understanding an utterance means encoding it into one’s own conceptual structure.

So far so good, he said nice things and represented me accurately. But then….

Schank does not believe that an individual’s entire base of conceptions can be explicitly extricated from him. He believes only that there exists such a belief structure within each of us and that if it could be explicated, it could in principle be respond by his formalism….

There are two questions that must ultimately be confronted. First, are the conceptual bases that under linguistic understanding entirely formalizable even in principle as Schank suggests and as most workers in AI believe? Second, are there ideas that, as I suggested, “no machines will ever understand because they relate to objectives that are inappropriate for machines?” ……

It may be possible, following Schank’s procedures to construct a conceptual structure that corresponds to the meaning of the sentence, “will you come to dinner with me this evening?” But it is hard to see — and I know this is not an impossibility argument, how Schank-like schemes could possibly understand that same sentence to mean a shy young mans’ desperate longing for love. 

I quoted parts of what Weizenbaum had to say because these were the kinds of questions people were thinking about in 1976 in AI. Weizenbaum eventually became anti-AI, but I always like his “dinner” question. It is very right-headed and it is the least we can ask of any AI-based TA or mentor. Can we build a program that understands what the student is feeling and what the student’s real needs are, so that we can give good advice? Good teachers do that. Why should future online teaching be worse than what good teaching is like today without computers or AI?

Do we actually have to do all this in order to build AI?

Could we simply build an automated TA/ mentor that did not do all that but still performed well enough to be useful?

These are important questions. Maybe Goel’s program did perform well enough to consider using it in MOOCs where there are thousands of students. I am not fundamentally interested in that question however.

Here is what I am interested in. Can we stop causing people to so misunderstand AI that every ELIZA-like program makes headlines and causes people to believe that the problems we were discussing in the 70’s have been solved?

The fundamental AI problems have not been solved because the money to work on them dried up in mid 80s. There are businesses and venture capitalists today who think they are investing in AI but really they are investing in something else.  They are investing in superficial programs that really are ELIZA on steroids. Would it be too much to ask people to think about what people do when they engage in a  conversation and build computer programs that could function  as an effective model of human behavior? I hope we can get people with money to start investing in the real AI problem again. Until we do, I will be finding myself on the side of Weizenbaum when when he was being critical of his user’s  reactions to ELIZA (for good reason.) We should start working on real AI or stop saying that we that are. There is nothing to be afraid of about AI, since hardly anyone is really working on it any more. Most “AI people” are just playing around with ELIZA again. It is sad really.

Weizenbaum and Colby were brilliant men. They were both asking fundamental questions about the nature of mind and the nature of what we can and cannot replicate on a computer. These are important questions. But, today, with IBM promoting something that is not that much more than ELIZA people are believing every word of it. We are in a situation where machine learning is not about learning at all, but about massive matching capabilities to produce canned responses.  The real questions are the same as ever. What does it mean to have a mind? How does intelligent behavior work? What is involved in constructing an answer to a question? What is involved in comprehending a sentence? How does human memory work? How can we produce a memory on a computer that changes what it thinks with every interaction and gets reminded of something it wants to think more about? How can we get a  computer to do what I am doing now — thinking, wondering, remembering, and composing?

Those are AI questions, not questions. They are not questions about how we can fool people.

Monday, May 9, 2016

Boredom spurs creativity; are computers or mobile phone owners ever bored?

Boredom matters. We need it. But, two sets of supposedly thinking entities are never bored: “smart” (deep learning) computers, and people who are attached to their phones (which is beginning to look like nearly everybody.)

A friend’s teenage son (who was coming over for some advice) rang my doorbell the other day. In the time it took me to open the door, he was already looking at his phone. When I am on the elevator in my New York apartment I notice that literally everyone is look at their phones during the ride. Sherry Turkel has pointed out that this behavior is killing conversation and she is right. But it is also killing something even more important: creativity.

Creativity depends upon many things but a key one is boredom. When you are bored your mind wanders. You do this weird thing called “thinking.”

I have begun thinking more about AI in recent months because of the incessant nonsense being written about what computers can or might do. So. let me ask a simple question. Is Watson ever bored? Do these “deep learning” machines get bored? It seems obvious that they don’t. Why not? Because, in order to be bored you have to have something you like doing, a goal you are pursuing, a problem you are interested in, or wondering about and are in some way prevented from solving. 

Wittgenstein said that all creative thinking took place  in the “three B’s”: bed, bath and bus. What he meant was, that that was the only time time there was no one else talking or distracting him and with nothing much to do, his mind wandered and interesting thoughts occurred.

When could this possibly happen in the life of young people who cannot stop looking at their phones? What is there to be bored with or bored about? If you are bored with a facebook post you just go to the next one. If you are bored with whats on TV each change the channel. If you have nothing to do you surf the web. No one sits quietly and thinks any more.

I find this very scary for two reasons. Our educational system is in such bad shape in party because we don’t allow boredom which means we really do not encourage creativity. There are answers to be memorize, books to be read, and test to be  taken. We aren’t actually expected have original thoughts in high school ever. (Unless a kid happens to have a really good teacher and more freedom than is typically allowed. 

Now computers. The very idea that AI is progressing is patently absurd. What would it mean to have a smart computer that didn’t on occasion have an original idea about something? How could a computer be smart if it didn't worry about things from time to time. Americans are busy worrying about a Trump-Clinton election. We talk about it. We wonder about it. That worrying looks like thinking. What computer would worry about this? How could a computer possibly worry about this? Does Watson worry?

Now, of course that is the real AI question and the kind I used to work on when AI was funded by people who thought AI was something other than “deep learning.” I asked myself  and my students how we could get a computer to have creative thoughts. One answer is that a computer would have be trying to figure things out in some way, be considering hypotheses about whatever it is trying to explain, then imagining alternative explanations, and then trying to invent one’s own. This is what creativity looks like.

Could a computer do that? Of course it could in principle, but it wouldn’t be the so-called AI machines we have now which are very good at counting and matching and searching., That kind of AI depends on being annoyed by a state of affairs and thinking you should be able to come up with some better answer and then putting yourself in the equivalent of a bathtub or a bed or any place where it is quiet and there are no distractions so you can let your mind wander.

Computers will not become creative (or bored) any time soon unless those who fund AI change their perspective.

What really bothers me is that people won’t be creative  either. Young people’s first thought these days is to post a picture of what themselves or what they are are looking at, rather than to think about the world around them.

Tuesday, May 3, 2016

AI is nowhere near working; let's think about what people can do that AI can't

When I started working in AI in the 1960’s, it wasn’t really one field, just a set of people trying to get computers to do some interesting things that we knew people were capable of doing. These days, unfortunately, AI seems to mean “deep learning” whatever that is, and stuff IBM talks about that uses the word “cognitive.”  

I have recently been thinking about some of the aspects of AI that I did not work on. (I worked on natural language processing, memory, and learning.)  I think there are things worth discussing about the other areas of AI that might shed light on what is really going in today’s so-called AI.

Let’s start with Face Recognition. It is clear that face recognition technology is pretty good. Facebook can tell when your picture has been posted by someone and can add your name to it. I am sure there are all kinds of surveillance technologies that make use of face recognition as well.

But, there is an aspect of face recognition that people naturally do, but computers cannot come close to doing today. I don’t mean to be political here, but my best example is recognizing Ted Cruz’s face. I can recognize him, but every time I see a man who is angry, mean, and just a little Satanic looking. Commentators say these things all the time, and I am not trying to comment on that; rather I am trying ask the question: what is it that we see when we look at someone and immediately distrust them and are slightly afraid of them?

To put this another way, when you are walking down the street and someone scary walks by, what is it that you see in his face that makes him seem scary?   I  have been running some experiments about how people react to talking head videos that we’ve captured of experts telling stories about their expertise. Every time someone looks at one of our videos they have an instant reaction to the human qualities of the person as well as to the story the person is telling. They like or dislike people in about ten seconds. What is it that they are seeing?

This is an interesting topic, but my point is about AI. AI is no where ready, no matter how well it does at face recognition, to tell us,  “I find this guy scary” or “distrustful" or “he seems to be lying”, even though people can do this all the time without conscious thought.

What does this tell us? It tells us that AI has a long way to go before it can do stuff that nearly any human can easily do.   Actually, any dog can do this. They too have instant reactions to a person. What are they seeing? This is the AI question. My guess is that Facebook even with its 100’s of AI people is not working on this problem and moreover, it doesn’t care. But it is a very important aspect of cognition. (Sorry, IBM, you don’t actually own that word.) Facebook is only working on the “you can count the pixels and pattern match” part of face recognition. When we feel attracted to someone, or we want to avoid them, we are using our innate ability to do a more subtle kind of face recognition.

I have this same problem with speech synthesis and speech recognition. I was riding in my wife’s car the other day and the navigation system  she was using told her to turn on “puggah” boulevard. We were in an area we know and both laughed out loud. The street is called PGA Blvd, which is the acronym for the Professional Golf Association. The program never heard about acronyms I guess. Later it told us to get on the ramp for W Palm Beach. Now a reader would think I am abbreviating west with the W, but I am not. The navigation system actually said “W.” My reaction was that this device is really stupid. How hard would it be to make an intelligent navigator with respect to speech synthesis? Well, apparently too hard for the company that made this one. (It would also be too hard for it to tell me about a new restaurant that I was passing and might want to check out, which is the kind of AI that I am interested in.) That is AI too, but it is not “deep learning,” so no one is funding it.

Which leads me to what I really wanted to talk about here, speech recognition. Someone said to me the other day that AI has made real strides in speech recognition. I laughed. Now, I realize people talk to Siri and other devices. And sometimes Siri “knows” what you are saying in the sense that it can find a response. As a way of pointing the real AI problem out to my friend, the next thing I said was:   “szeretlek nudunuca”. To which he responded “huh?” I said it again. He said he didn’t understand. I asked if he could tell me one word that I said. He said “no.” I said “can you even report a part of what I said?’ He said “no.” It all sounded like gibberish to him. Of course it did. I was talking Hungarian. When someone speaks an unfamiliar language you cannot hear where the word breaks are, and you cannot even decipher the sounds. This is because human speech recognition involves having heard everything before and understanding the context in which the spoken words said belong.  

It is difficult for people to understand a sentence that is out of context.  What is a normal response to a completely unexpected sentence? People generally have to be listening for something in order to understand it. Understanding involves guessing about what someone is likely to say. Those guesses are made on the basis our knowledge of each other and of the possible things we are or might talking about. To do that right in AI, we need to determine intentions and motivations and we need to have a model of the person we are talking to, including what they know and what their interests are.

The other day someone who I play softball with (who often asks me questions) asked me “what is Zion?” I had to ask him to explain what he was actually asking about. I heard the words, but had no idea what he was trying to find out.  After a bunch of sentences from him I got what the question was about. I didn't have any trouble with the words, but I had absolutely no idea what he wanted to learn from me. Siri and the others would not be able to have that conversation with him because there is no AI there. Apple, Google and the rest don’t care about that. It is the pretense of AI that seems to interest them.

We are very far from a computer that can do the things I’ve just discussed. AI will be hyped as much as IBM’s marketers and others choose to do it in an effort to make money. As for me, I would prefer that they actually worked on AI instead of trying to convince everyone that AI is already here.