"John hit Mary" and other AI problems

With articles being written about AI constantly now, I feel it is time to think about the basics. How do you get a computer to understand a simple sentence such as John hit Mary?

When I first started working on computational linguistics (as it was called then), the linguists made clear that they thought this was easy to do on a computer. You just used a syntactic parser and identified the noun phrases and the verb phrases.

I thought this was absurd, just as absurd as the idea that AI is coming tomorrow to eat us all.

To explain why, let’s discuss this sentence. What happens when you hear it? You react in some way. You might think that this was spousal abuse and that the police should be called. But then, I could tell you that John is 5 and Mary is his mother. Then you might wondered if and how Mary punished him. Or, I could tell you that John is 50 and Mary is his mother. Now you are wondering about the police again and also about what is wrong with John. Or, I could tell you that Mary is 5 and John is her father and you would be wondering about his parenting skills.

Absent of any of this information you make assumptions. I could ask you what Mary was wearing or what color her hair was and you might very well have an answer, or at least a guess. We comprehend through visualization and imagination. No sentence makes full sense out of context and we rarely have the full context, so we imagine it. People are constantly figuring out the parts they don’t know for sure. We make mistakes all the time. That is how human comprehension works. Did I mention that John and Mary were each driving their own cars at the time? Did I mention that John is a blackjack dealer and Mary was playing blackjack? Perhaps John is a baseball pitcher and Mary was the batter. Maybe they are both boxers.

Sentences don’t mean much out of context but two things are true:

1. we never have the full context so understanders make inferences, draw pictures in their minds and attempt to do the best they can

2. computers, in order to do this effectively, would have to have what people have: a model of the world. 

When you hear about all the AI programs being worked on today it is safe to assume that they are not even thinking about building complex world models.

For fun, I typed “John hit Mary” into Google. The first thing that comes up is an article on empathy that discusses some of what I have been saying here. The second thing that comes up is an excerpt from a book of mine (Explanation Patterns) which discusses the belief structures underlying the comprehension of such a sentence.  

Since the media and many companies would have se believe that have solved the natural language problem, let’s consider a real example of “John hit Mary” . This is from the New York Post (July 25, 2016);

Cops smashed their way into Hollywood star Lindsay Lohan’s posh London flat after a furious bust up with her Russian lover.

Police were called in as Lohan, 30, suffered a meltdown on the balcony of her Knightsbridge apartment with boyfriend Egor Tarabasov, 22 – claiming she had been attacked.

Waking up neighbors, the A-lister shouted, “He just strangled me. He almost killed me.”

In footage taken by a neighbor she could be heard begging for help outside of her $4.2 million home at 5 a.m. on Saturday morning.

Shouting her name and address across the street, she screamed, “Please please please. He just strangled me. He almost killed me. Everybody will know. Get out of my house.”

She added, “Do it. I dare you again. You’re f—ing crazy. You sick f—. You need help. It’s my house, get out of my house.”

She was also heard shouting to Egor — who was also near the balcony, “I’m done. I don’t love you anymore. You tried to kill me. You’re a f—ing psycho,” before adding. “We are finished.”

She added, “No Egor, you’ve been strangling me constantly. You can’t strangle a woman constantly and beat the s— out of her and think it’s ok. Everybody saw you touch me. It’s filmed. Get out! Get out.”

Ten minutes later police arrived after receiving reports of a “woman in distress” – forcing their way into the property – only to find it empty after going inside.

They said no crime was committed and no arrests were made.

How would a computer understand this story? It would need a better model of the world than I have. I don't know much about Lindsay Lohan except that she was a popular child actor who now seems to be in trouble quiet often.

Reading this story, I wonder what is wrong with her. Why does she make such bad choices? How is a 22 year old Russian the right man for her? 

Then, also, I wonder why no arrests were made. Was she making it all up? And how about “you have been strangling me constantly?” Really? Why would someone put up with that? Sometimes, I know, poor women put up with abuse because they have nowhere to go. But isn’t she the rich one?

So, when I read this, I wonder many things about what is wrong with this woman, why her life went so wrong, why someone isn’t helping her, and what parts of this story have been left out.

A computer would need a deep model about why people do what they do, better than the one I have, because I am even having trouble with wondering why this is news. (Of course, it is the New York Post.)

When computers can tell me what the real issues are here and be able to enlighten me in some way about the  questions I asked, then I would be impressed. Not afraid this AI, simply happy to know that a computer could explain stuff to me that I don't have a good world model for, and hence don’t understand very well. It is the building of complex world models about why people do what they do that underlies all understanding. The AI being worked on today doesn't even attempt to solve that problem.

5 questions about human intelligence that make clear AI is far from here yet

I have some questions for you:

1. How many windows were in the house or apartment in which you lived when you were ten?
2. Can you name all 50 states? (For Europeans: can you name every country in Europe?)
3. What was served at your birthday party when you were 13?
4. When you came back from your first trip abroad, how did you describe the experience to your friends?
5. What was the most difficult interaction you ever had with a teacher and what did you learn from that experience?

Why am I asking these questions? The popular world has suddenly become obsessed with AI. Venture Capitalists have become obsessed with funding AI companies. I thought it might be helpful if we discussed I (as opposed to AI) a little bit. You can’t really expect an AI to take over the world if it isn’t intelligent. Since the media are so concerned with this impending take over, I thought I would take a shot at explaining some aspects of intelligence in humans and the properties of human memory upon which it relies that AI will have to emulate to be intelligent.

So, question 1, how does one answer it? Actually the answer is pretty simple. You need to take an imaginary walk around your dwelling and count the windows. I always used this example in my AI classes. Why? Because taking an imaginary walk around a house requires a visual memory. We can remember what things looked like, imperfectly, typically, and can find the answer. There is nothing to look up. No data to search. No “deep learning to be had.” You simply have to look. But how simple is that? Can we create a computer that can walk around its own prior visual experiences? Possibly. But the computer would have to have remembered what it saw, not in terms of pixels but in conceptual terms. (“There was a green couch in the living room, I am pretty sure.’) So, memory is visual, but it is also reconstructive. We figure the couch had to have had an end table nearby but we don’t remember it, so we imagine it and attempt to reconstruct it. People get into arguments with family members over this kind of stuff because our memories are imperfect and we reconstruct in idiosyncratic ways. An AI would need to be able to do that. (Fight with its siblings? Yes.)

I asked question 2 in my classes every year. (Former students do you remember that? Maybe you do and maybe you don’t. Can you remember why I did it?) I did it because I was trying to explain the difference between recognition and recall memory. I can’t recall a student who could actually name all 50 states. (There my have been one or two). Mostly they got 47 or 48. They usually left out Utah or Idaho or Arkansas. Why? When I pointed out the states they had missed, no one every said: I never heard of that state. They knew the names of all 50 states but in order to name them they didn’t search the web to find the list. Modern AI’s would have that list. But modern AI’s aren’t really intelligent and don’t behave the way humans do.) They can just search lists. How do people do it? They “walk around” a map that they can visualize. They go down the East Coast and they go up the West Coast. They rarely miss any of those states. It is those darn middle states that cause all the trouble. Why? Because the maps that we have stored in memory are imperfect. Memory is very important in human intelligence but we are kind of bad at at. Does this mean AI’s will beat humans at memory tasks? They might. They probably could name all the states but they would do it very differently. They could win Jeopardy but not by doing what people do wen they recall information. Does this matter? Yes it does. I am getting to why.

Question 3. Why would anyone remember what one ate at a birthday party many years ago? You might not. Part of human memory is its ability to be selective about what it remembers. Not all experiences are equally important. We need to learn from the important ones and disregard the unimportant stuff. Can AI do that? Not that I know about. “Importance” implies that one has goals. These goals drive what we pay attention to and what experiences we dwell upon as we grow up. Oh, but modern AI’s don’t grow up. They just search, and store, and search some more. They don’t get wiser from each experience. And they don’t reconstruct. I have no idea what the food was at my 13th birthday party, but that was a big occasion in my world so I can guess. I really would guess badly because the food was not the issue, the party was. (And, I have pictures, but only of the bread and the cake.) My memory helps me figure out answers, but it does not provide them. My memory is full of experiences that I have to re-interpret every time. That is what intelligence is based upon, faulty memory. So, modern AI can make better memories perhaps, but of what — words in texts? My memory is based upon emotions.That was a big day for me. I remember cousin Joanie dancing. (Or was it my girlfriend Phyllis?). I remember my grandmothers kissing me. I remember my mother’s yellow dress. Memory is like that. (And since I am male it is not shocking that I remember the females, who always held (and still do hold) a fascination for me.)

Question 4. My first trip abroad, which lasted about a month, has maybe five salient memories. One was watching my mother do business in Austria and noticing that she had failed to notice something her competitors were doing that was hurting her. A second was driving around some of Eastern Europe by myself, a drive which included me passing a farmer in a wagon in Yugoslavia and feeling him hit my car with his horsewhip. (Maybe I wasn’t supposed to pass him.) A third was meeting a girl on the plane from Vienna to Tel Aviv simply because I asked her a question (in English of course) and she was ecstatic to find someone else on the plane to talk to. (Our relationship lasted all of two weeks, but I remember it.) A fourth (this was 1967) was seeing the Israelis already building settlements on the West Bank and me wondering how exactly doing that would lead to peace.) The fifth was my visit to Venice where I was hosted by a cousin who tested my “American crudity” by asking me to eat spaghetti, assuming that I would do it wrong and her being disappointed when I didn’t. (I grew up with a lot of Italians in Brooklyn.) Why am I telling you this? Because this trip lasted a month. I can remember a little more about it but not a month’s worth of stuff. I remembered stuff that caused me to learn something important  about business, about how to meet women, about international politics, and about things I still don't understand —e.g.  the farmer with the whip.) We learn from experience. Any serious AI program would have to do the same. Too bad what we mean by AI today isn’t even close to what I am talking about.

The last question is obvious. A good teacher makes you think. I had plenty of those. I also had one who hit me. I didn’t learn much from that except to stay away from her. As I write this, I am on the way to the 90th birthday party of my PhD thesis advisor (Jacob Mey.) All my interactions with him were difficult. From each one I came out wiser. I learned from being criticized and I learned from being told I was wrong. We argued. I learned. When AI programs do that, we will have AI. Until then, not so much.

Argumentation, goals, emotion, visualization, imagination, and reconstructive memory. Stop worrying about current AI programs. Or, start worrying about them. Because they sure aren’t doing those things.

Six things computers (and people) must do in order to be considered intelligent

We hear a lot about AI these days, most of it pretty silly. It seems all to be about answering questions by key word matching and finding ads based on search. To me, AI has always been a field at the centre of which was intelligence. Here, I will list 6 things that most intelligent people can do, that no AI program can do. While Hawking and Gates are very afraid of AI, I am very afraid that no one is working on the right problems in AI any more.

1. People can make predictions about the outcome of actions

So, I could ask a person: What do you think will happen if we keep having elections for President when a large chunk of the population doesn’t like either candidate?

This would start a conversation about the current election. It might lead to an argument. It might lead to a solution. Type this in to Google or Siri or Watson and see what you get. Hint: you get newspaper articles that match on some of the words. 

Conversation is a hallmark of intelligence. Any AI system must be able to have a conversation about a complex topic. All the “deep learning” that is going on is not focusing on that very simple test of intelligence.

2. People build a conscious model of a processes in which they engage

Here is something someone might say: I keep hearing about global warming. Should I be fearful? Isn’t this just threat for those who live in coastal areas? The climate has always been changing.

This question calls for someone who has a model of global warming both now and historically to respond to it. What “AI” could do that today? Who is even trying? (Hint: it is very hard.)

3. People find out for themselves what works and what doesn’t by experimenting from time to time.

Something someone might say: I wonder how likely it is that I would get a speeding ticket if I went 120 on I-95.

A reasonable response to this might be: Where on I-95? or Why would you want to take that risk? 

Find me the AI group that is working on helping people figure out how things will turn out if they try something new.

4. We are constantly evaluating things. We attempt to  improve our ability to determine the value of something on many different dimensions

One might say: I think she is in love with me. How do I know for sure?

Typically people respond to such sentence with stories of their own lives, of love that went right or went wrong. Find me a computer that tells you a story when you are worried about something in your own life. (I did work on this problem and still do. But you can be sure Microsoft’s AI group isn't working on it.)

5. People try to analyze and diagnose problems they have to determine their cause and possible remedies.

For example: My business has had flat earnings for two years now. Should I be worried? What can I do about it?

A normal person would try to find an expert to ask these questions of. I would like to have a computer expert to ask these questions of. Google responds with a four year old article from Atlantic magazine about buying a house:

http://www.theatlantic.com/business/archive/2012/03/we-wish-like-hell-we-had-never-bought-voices-from-the-housing-crisis/253888/

I assure that any natural language processing program that Google is working on would not fare much better here.

6. People can plan. They can do needs analysis as well as acquiring a conscious and subconscious understanding of what goals are satisfied by what plans

Example: I am thinking of moving. But I am wondering what will happen to my relationships with the people who live near me.

When computers have stories to tell and can relate an experience or concern that a person has, to something it knows about and start a reasonable conversation with them then we will have AI. I would not be afraid of that AI. I would welcome it. But, unfortunately no one is working on this. Companies are saying AI constantly we and building up expectations in people that will not be satisfied unless and until the so-called AI companies work on these six problems (and many more.) 

These six problem underlie intelligence, artificial or otherwise. Time to think about intelligence and not Markov Models that make search better.

To summarize: Intelligent people have memories. They augment those memories through daily experiences and human interactions. They don’t have knowledge stuffed into their memories, instead they learn through attempting to achieve goals they inherently have and finding that the plans they tried need to be adjusted. They get help in the form of stories from other humans, told just in time. When computers can do all this, we will have AI. Right now, we have a lot marketing and hype.