Artificial Intelligence
Gongs and Parking Tickets
Please click for a bigger image
I have twin girls who were three years old a couple of days ago (23rd September 2019). One of them, Lily, is very loveable and takes after her mother. The other one has a mass of curly hair and a very independent nature (I know).
In the hall of my house we have a gong and on the top right hand side there is a hook from which a mallet (the correct name for the thing you hit the gong with, apparently) normally hangs - see photo above.
When Lily was 18 months old I walked into the hall and found the mallet lying on the floor (everything ends up on the floor if you have kids). I picked it up and walked to the gong and hesitated - I was looking for the hook to hang it on - something I have not done for many years. At that point Lily walked up and pointed to the hook with her finger. Not vaguely but so that her finger tip was an inch from the hook.
Frankly, I was stunned - neither of us had said a word - how did she know what I was hoping to do? So let's examine that from a computing perspective.
She sees me pick up the mallet and walk to the gong and hesitate. She has never seen me carry the mallet or even touch it because I just don't. But she instantly deduces from my hesitation and facial expression that I want to hang it up and don't remember where the hook is. I cannot even conceive of a computer program that could do that - never mind write one.
If you think this is easy, you have not understood the problem. Your laptop can work out one million square roots per second. I can write code from first principles to do that in five lines. It is a trivial problem. To deduce somebody's intentions from their facial expression and actions is a World Class problem. Maybe even a Universe Class problem. Astounding.
So why is it so difficult? Computers can beat the world champion at chess and now Deep Mind has beaten the world champion at "Go". So, why can't they do simple things like explain what is happening when you show them a photo? Or answer a simple question "if I balance a bucket of water on top of the door what will happen when somebody walks through the door?"
BECAUSE THEY CAN ONLY DO WHAT A HUMAN BEING HAS WRITTEN CODE FOR THEM TO DO. They are totally clueless when confronted with something new and different that is not in the computer program.
Now over the years many people have claimed that their computer program "learns to do things". Really? Deep Mind is the latest outfit to claim this but let's examine some uncomfortable facts. Deep Mind lost £250m last year. How do you lose that much money? The only way you can do that is to have maybe 1,000 coders all earning £250,000pa.
So when they say "the machine taught itself to play Go and beat the world champion" what they actually mean is: "a team of hundreds of the world's best minds spent two years writing millions of lines of code that enabled the computer to learn to play Go". Did Lily have a team of hundreds of the world's best minds writing code to enable her to deduce my intentions from my facial expression. No, she did not.
Here is another example:
Last week we parked in Alrincham and the girls took it in turn to put the coins in the machine - a pound in 10p and 20p coins. Then Lily reached into the machine extracted the ticket, detached the adhesive part and stuck it on the back of the ticket correctly, opened the door of the car and stuck the ticket the right way up on the door window so it was visible.
I was astonished. How on earth did a child not yet three years old do that? We have never taught her and even adults struggle to get the ticket stuck on correctly (I know I do). She had picked this up just by watching. Could Deep Mind do that? Of course, after several hundred programmers have spent two years writing the code to enable it to "learn to walk down the street and handle parking tickets"
There is no computer program on earth that could even remotely do that - including MOPEKS (yet). That is just stunningly difficult - another Universe Class achievement.
MOPEKS
In 1986 while living in Atlanta, Georgia I read Douglas Hofstadter's classic book, Godel, Escher, Bach: An Eternal Golden Braid. After I had read it I vowed to work out how the human brain operated.
So from 1986 until 2013, this was my main occupation. My business just ticked over in the background. By 2008 I was making progress and then spent 5 years and 5,000 hours (three hours every morning with no exceptions) writing computer code to turn those ideas into a prototype computer program written in a mixture of VB6, C and ASM
In 2013 I published the results as a book Mopeks - a Route to Intelligence and a website MOPEKS.
Am I right? Yes, I genuinely believe I am. I have no problem in admitting I am wrong on occasion. If on reflection I thought my approach was fundamentally flawed, I would just say so "yes, I got it wrong". So, why has MOPEKS not taken over the world?
The short answer is that to turn MOPEKS from an extremely limited prototype into something that can actually solve real life problems is immensely difficult - there are a lot of World Class problems in the way. But in 2020 I intend to start work on Phase II and see if I can make a small part of it fully functional and then we will see - but don't hold your breath - I may be gone some time (thank you Captain Oates)
Conclusion
You are NOT going to lose your job anytime soon because of AI. In the 1980s, countless media hacks made a comfortable living predicting the "paperless office". We are still waiting. These days they make an even better living by saying "you are going to be replaced by an AI program". Don't worry - you are not.
Why not? Because this is just helpful technology and nothing else. "A computer program can now detect whether a skin melanoma is cancerous better than a doctor can", I read last week. So what? An X Ray machine can detect a bone fracture better than a doctor can. And blood tests can diagnose your illness "better than a doctor can". And a hundred other chemical tests are "better than doctors". But we still need doctors to cure the problem once it has been found.
The same applies right through society - AI in its present formulation is just a tool to solve a very precise problem. In 1965 I was programming mainframes that were going to take over the world and make people unemployed. They didn't and they aren't.
But if anybody (me maybe) does produce a truly versatile intelligent program then that is the singularity and we are all totally screwed. As soon as it emerges it will infect every internet connected device in the world so it is totally impossible to kill. It will then take total control of all communications and weapon systems and will have total de facto control of the entire world.
The President of the USA will find that he is totally powerless as will all other world leaders. Then we had better hope for the best because that is all we can do. Bruce Willis and Jason Statham are not going to save you. But if you can offer me special terms on my next deal I will see what I can do to help ...
Bob Cory
Modified on 27/10/2019 at 11:33:54 by ℗ Bob Cory