Previous posts in this discussion:
PostDevelopment of Artificial Intelligence: 1970s and Today (Istvan Simon, USA, 05/14/18 9:51 am)
I got my PhD at Stanford in 1977. Stanford has always been one of the most important centers for Artificial Intelligence (AI). I in fact took some courses in AI from the late John McCarthy, one of the pioneers of AI, and the inventor of LISP. In the 1970s and later I pretty much had the opinion that John Heelan expressed about the limitations of Artificial Intelligence (May 4th). In those days AI was quite far from anything close to human intelligent behavior, There were some successes in restricted parts of AI:
1. Expert Systems were indeed comparable to human performance and even superior to humans in certain areas. For example there was an expert system that could advise on the proper use of antibiotics for doctors, and it was expert indeed and useful. Similarly, Dendral was a success about Chemical Formulas.
2. Computers beat the best human players in checkers, but chess and Go were much beyond the capabilities of computers to play at the best human players level in those days.
The best computer chess program was Belle. Belle did not play chess with any AI techniques at all. It was written by Ken Thompson, the main inventor of Unix. Belle had a recording of all known openings, and would play according to the book, much like human players do too in the opening. When the game deviated from the book, Belle would resort to an alpha-beta search, which is essentially equivalent to a program that chooses the best move available, assuming that the opponent would then choose the worst for us, then choosing the best move for us, assuming that the opponent would choose the worst for us, and so on. This is called a min-max strategy. It works like this: Suppose from a given chess position we construct the tree of all possibilities, looking ahead say 10 moves. Say it is our turn to move, so moves 1, 3, 5, 7, and 9 would be our move, and moves 2, 4, 6, 8, and 10 would be moves by our opponent. We then would have an enormous number of positions that could be reached from the current position by any sequence of 10 moves. Each of these positions would be then given a number of its worth to us by a "static evaluator." A static evaluator would use simple rules for chess to evaluate a position, described in all chess books. For example, it would use the values of the pieces still on the board, 10 for pawn, 30 for knights and bishops, 50 for rooks, 100 for queens and -10 for our opponent's pawns, -30 for our opponent's knights and bishops, and so on. It could have also some value assigned to other desirable features, like dominance of the center of the board and so on. So the static evaluator is a relatively simple algorithm that computes the value to us of any chess position by such simple rules. The best move for us would be the sub-tree of this enormous tree that would produce the best position for us, by the mini-max strategy. Belle would choose that move.
i emphasize that this is not at all how humans play chess, but at the time this was the best computer chess program. It was understood that if computers could see far enough ahead, which was limited only by the speed of processing power available, they could beat any human player by this simple strategy. But in the 1970s this was simply not possible. However we all know that since then Big Blue defeated the human world champion in chess, so much progress has been made in computer's abilities to play superior chess.
3. Stanford had an AI project called the hand-eye project, which was the foundation of intelligent robots today. Robots indeed are better than humans in assembly work, and they do not strike, need health benefits, retirement pay, and they work 24 hours a day without complaint, so clearly they are very useful.
4. Computers could prove theorems, but when compared to humans in general they were pitiful, except in certain restricted fields, where humans also do not have much insight or intuition.
5. Machine speech recognition was primitive, and computer translation was almost useless when compared to a human translator. Computer "understanding" of language was primitive and unsophisticated.
The above was a fair summary of computer AI program performance in the 1970s. But much changed in the years since, and computers definitely got much better at all of the above. Machine translation for example today is very usable, and computers beat the best humans in chess, and even in Jeopardy. So AI has much improved, and I think that John Heelan's points today are no longer valid.
JE comments: Bravo for Istvan Simon's optimism. On the back burner of my anxieties is the expectation that machine translation will render it pointless to actually study a language. Perhaps I'll be retired by then. And at least we'll still have WAIS to keep me out of mischief.
Thoughts on AI Today; UK Emergency Health Services
(John Heelan, UK
05/15/18 7:24 AM)
I bow to Istvan Simon's greater knowledge of the AI world as it is today (14 May). Not only am I hopelessly out of date technically, despite a close friend also having a PhD in cybernetics, but also when talking to today's more advanced practitioners in the computing world (some in my family), I realise that my level of knowledge is Stone Age in computing terms.
My worry is that we are becoming too reliant on the promises of technology that are often not forthcoming. A good example in the UK is the emergency health services. If unwell I am encouraged to ring 111. However, if I do so, I am confronted with a retrained Call Centre clerk attempting to diagnose my level of emergency from a script on his/her screen.
Another emergency service if the Fire Brigade. Locally there is confusion when a "blues and twos" shout is directed to a village called Niton (pronounced "Nye-ton" when it should have gone to similar-sounding Knighton. Local dispatchers have learned to ask "is that Niton or K-nighton"? Would AI be able to ask the question of understand the spoken answer from somebody whose house was in danger of being consumed by flames?
My wife recently had to call an ambulance for an elderly man who had fallen and damaged his leg. The dispatcher was not even aware that our village was on the Isle of Wight, delaying the arrival of the paramedics for an extra half an hour.
JE comments: The UK 111 is our 911. This could be the start of an interesting comparative discussion. What are some WAISer experiences with emergency help lines? How about the wait times in different nations and regions? Several weeks ago in Delaware, my Aunt Doris needed emergency assistance. The 911 paramedics arrived in about 5 minutes. Bravo to them.
Emergency Services in Russia
(Istvan Simon, USA
05/22/18 9:55 AM)
I share John Heelan's concerns regarding an over-reliance on technology (most recently, May 15th). Whether or not the UK's 111 system is a good move or not, I cannot say, and it is quite possible that it is a blunder. Only time will tell.
Recently I saw a report on emergency services in Russia, in which they specialized people for certain kinds of emergency with the idea that it would make the system more responsive and efficient. I am not sure if this was entirely a documentary, or somewhat fictionalized. It followed one very conscientious emergency medical technician, who refused to follow that new rules and it seemed for very good reasons. He risked being fired, yet he did not follow the new manager's orders.
In this program all emergency personnel viewed the move as stupid and detrimental to emergency quality of service. Yet they were ignored by management, The rules that John mentioned seemed somewhat similar to the measures that appear in the program in Russia, though they did not seem to have any AI program involved. But the gist of the new rules was that the dispatcher was supposed to select the correct crew for each emergency, and that an emergency call was supposed to be concluded in a limited amount of time. If the emergency was not resolved within the allotted time, the crew was supposed to leave and another crew dispatched to the scene to continue their work. This was considered particularly stupid, and at odds with the needs of patients.
The program John Heelan describes in the UK has some similarities to the Russian one, namely the dispatcher in both systems making complex decisions that she or he is ill-equipped to make, whether assisted by an AI program or not.
JE comments: I hope Cameron Sawyer will weigh in. How could it be more efficient to send a second, "replacement" crew to an emergency? For starters, think of the learning curve involved with each incident.
- Emergency Services in Russia (Istvan Simon, USA 05/22/18 9:55 AM)