Wednesday, February 08, 2017

7 myths about AI

Big Blue beat Kasparov in 1997 but chess is thriving. AI remains our servant not master. Yet mention AI and people jump to the far end of the conceptual spectrum with big-picture, dystopian and usually exaggerated visions of humanoid robots, the singularity and the existential threat to our species. This is fuelled by cultural messages going back to the Greek Prometheus myth, propelled by Mary Shelly’s Frankenstein (subtitled The Modern Prometheus) to nearly a century of movies from Metropolis onwards that portray created intelligence as a threat. The truth is more prosaic.
Work with people in AI and you’ll quickly be brought back from sci-fi visions to more practical matters. Most practical and even theoretical AI is what is called ‘weak’ or ‘narrow’ AI. It has no ‘cognitive’ ability. I blame IBM and the consultancies for this  hyperbole. There is no ‘consciousness’. IMBs Watson may have beaten the Jeopardy Champions, Google’s AlphaGO may have beaten the GO Champion – but neither knew they had won.
The danger is that people over-promise and under-deliver, so that there's disappointment in the market. We need to keep a level head here and not see AI as the solution to everything. In fact, many problems need far simpler solutions.
AI is the new UI
AI is everywhere. You use it every day when you use Google, Amazon, social media, onine dating, Netflix, music streamming services, your mobile and any file  you create, store or open. Our online experineces are largely of AI driven services. It's just not that visible. AI is the new UI. However, there are several things we need to know about AI if we are to understand and use it well in our specific domain, and in this case it is teaching and learning.
  1. AI is ‘intelligent’
  2. AI is all about the brain
  3. AI is conscious
  4. AI is strong
  5. AI is general
  6. AI is one thing
  7. AI doesn’t affect me
1. AI is not ‘intelligent’
I have argued that the word ‘intelligent’ is misleading in AI. It pulls us toward a too anthropomorphic view of AI, suggesting that it is ‘intelligent’ in the sense of human intelligence. This, is a mistake as the word ‘intelligence is misleading. It is better to see AI in terms of general tasks and competences, not as being intrinsically intelligent, as that word is loaded with human preconceptions.‘intelligence’
2. AI is not about the brain
AI is coded and as such, most of it does not reflect what happens in the human brain. Even the so-called ‘neural network’ approach is loosely modelled on the networked structure of the brain. It is more analogy that replication. It’s a well work argument but we did not learn to fly by copying the flapping of birds’ wings and we didn’t learn to go faster by copying the legs of a cheetah – we invented the wheel. Similarly with AI.
3. AI is not cognitive
IBMs marketing of AI as ‘cognitive technology’ is way off. Fine if they mean it can perform or mimic certain cognitive tasks but they go further, suggesting that it is in many senses ‘cognitive’. This is quite simply wrong. It has no consciousness, nor real general problem solving abilities, none of the many cognitive qualities of human minds. It is maths. This is nit necessarily a bad thing, as it is free from forgetting, racism, sexism, cognitive biases, doesn’t need to sleep, networks and doesn’t die. In other words AI is about doing things better than brains but  by other means.
4. AI is weak
There is little to fear from threatening independence and autonomy in the short term.  Almost all AI is what is called ‘weak’ AI, programmes, run on computers that simulate what humans can do. Strong AI is the idea that it actually does what the brain does. My own view is that we are very firmly at the ‘weak’ stage but that the distinction is actually on a spectrum, like cool to hot. That’s not to say that ‘strong’ AI is not on its way, just that it’s not here yet.
5. AI is narrow
AI applications do specific things well and general things badly. They play chess and GO well but that piece of AI does very little else.  That is nit to say that AI will not get to the position of being a general problem solver. It will just atke time. We can see, from driverless cars, that an array of sensing, decision making and learning software can do remarkable things when working in tandem.
6. AI is not one thing
Far from being one thing, AI is many things. There are also many different approaches to AI. This is well covered in Pedro Domingos’s book The Master Algorithm, with chapters on Symbolists, Connectivists, Evolutionists, Bayesians and Analogizers. From the learning perspective, there’s machine learning (supervises and unsupervised), reinforcement learning and  and deep learning. Other ways of looking at AI is through areas of problem solving, such as NLP (Natural Language Programming) or robotics. There are many other ways of slicing the AI cake. The important point is to see it as a set of very different technologies or tools that solve problems.
7. AI doesn’t affect me
In practice AI is very good, often better than humans, in very narrow applications and domains. The most pervasive example is Google, which is brilliant at searching for relevant links and information – whether it be websites, scholarly papers, images, videos, audio, maps and so on. This has revolutionised how we store, access and use media and knowledge. It was a huge pedagogic shift. Another specific, but immensely powerful use, is in buying, where Amazon’s algorithmic power constantly recommends and shapes our buying habits – especially in books. Then there’s social media, where the algorithmic power of personalised news shapes out timelines on facebook or Tweets. On top of this are the algorithmic recommendations on Netflix and many streaming music sites. AI is the new UI. Most of our online experience is shaped by AI, as it gets to know us better and delivers a better service. We live in the age of algorithms.
AI is the wonder of our age. Something so exciting that it tends to produce extreme reactions in commentators. This is fine but we must, for the moment not fall into dystopian or utopian visions of what it can achieve. We must be realistic. That is not to say we should be complacent. We have already seen the power of AI to transform online services, automate factories and seriously impact employment. Political shocks, like Brexit and Trump can, I believe, be partly attributed to this phenomenon. We must therefore be vigilant on regulation and its political consequences. 

1 comment:

Anonymous said...

Donald, I am sure you have seen this from 2015: Won't be long before self-consciousness is a reality, then what?