The Rise of AI and the World of Machine Learning Peter J Bentley - PowerPoint PPT Presentation

The Rise of AI and the World of Machine Learning Peter J Bentley

What is AI? Robots? Magic software? Terrifying Super Intelligences? Machine Learning?

1951 This is a maze-solving machine that is capable of solving a maze by trial-and-error means, of remembering the solution, and also of forgetting it in case the situation changes and the solution is no longer applicable. …Now I would like to show you one further feature of the machine. I will change the maze so that the solution the machine found no longer works. By moving the partitions in a suitable way, I can obtain a rather interesting effect. In the previous maze the proper solution starting from Square A led to Square B, then to C, and on to the goal. By changing the partitions I have forced the machine at Square C to go to a new square, Square D, and from there back to the original square, A. When it arrives at A, it remembers that the old solution said to go to B and so it goes around the circle A, B, C, D, A, B, C, D …. It has established a vicious circle, or a singing condition. - A neurosis - Yes - It can't do that when its mind is blank, but can do it after it has been conditioned? - Yes, only after it has been conditioned. However, the machine has an antineurotic circuit built in to proven just this sort of solution -It doesn't have any way to recognise that it is "psycho" it just recognizes that it has been going too long? -Yes. As you see, it has now gone back to the exploring strategy.

Presentation of a Maze-Solving Machine Claude Shannon 1951

1950 I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted. I believe further that no useful purpose is served by concealing these beliefs. The popular view that scientists proceed inexorably from well- established fact to well-established fact, never being influenced by any unproved conjecture, is quite mistaken. Provided it is made clear which are proved facts and which are conjectures, no harm can result. Conjectures are of great importance since they suggest useful lines of research.

Computing Machinery and Intelligence A. M. Turing 1950

1945 In analyzing the functioning of the contemplated device, certain classificatory distinctions suggest themselves immediately. First: Since the device is primarily a computer, it will have to perform the elementary operations of arithmetics most frequently. These are addition, multiplication and division. It is therefore reasonable that it should contain specialized organs for just these operations... a central arithmetic part of the device will probably have to exist and this constitutes the first specific part: CA. Second: The logical control of the device, that is the proper sequencing of its operations can be most efficiently carried out by a central control organ ... this constitutes the second specific part: CC. Third: Any device which is to carry out long and complicated sequences of operations (specifically of calculations) must have a considerable memory ... this constitutes the third specific part: M. ...The three specific parts CA, CC and M correspond to the associative neurons in the human nervous system . It remains to discuss the equivalents of the sensory or afferent and the motor or efferent neurons . These are the input and the output organs of the device.

First Draft of a Report on the EDVAC John von Neumann 1945

AI has always been inspired by biology From living systems we discover what might be possible. AI enables computers to do what they currently cannot. (Once computers can do it, it’s no longer AI)

What can’t we do with computers? • We can classify and predict from data. • It’s a fashionable area in business and research – we develop new machine learning methods to solve significant problems in business. • But predicting the behaviour of complex systems is still tricky.

Can we predict the behaviour of developers?

Predicting behaviour of App Stores App Store Developer User App builds and downloaded by uploads

Russia Canada N=278 N=508 UK N=271 USA France Japan N=299 China Germany N=232 N=215 N=514 Mexico N=255 Spain South Korea Italy India N=245 N=260 N=258 N=261 N=344 Brazil N=430 Australia N=278 Cyprus, Malaysia, Belarus, Ukraine, Colombia, Costa Rica, Indonesia, Vietnam, Sweden, Guatemala, Kazakhstan, Singapore, Chile, Puerto Rico, Thailand, Argentina, El Salvador, Peru, Philippines, Croatia, Ecuador, Greece, Norway, Panama, Paraguay, Romania, Austria, Belgium, Bolivia, Caribbean, Dominican Republic, Fiji, Ghana, Honduras, Ireland, Ivory Coast, Kyrgyzstan, Mauritius, Netherlands, Pakistan, Poland, Portugal, St. Vincent, Switzerland, Taiwan, T urkey, Uruguay, and Venezuela.

(a) (b) ! An epidemic curve for a good app resulting from Mass The spread of the excellent infectious app through the user network using Exposure in an example run. (a) the Mass Exposure strategy, and (b) the Enhancing Mode of Transmission through New Apps Chart strategy.

Can we predict the behaviour of cancer?

Cancer Most conventional treatments are old. They are based on a reductionist approach. The effect of treatments on tumours is not properly understood.

Intestinal Crypts- Normal Case

Dynamical tracing at genetic scale Day 1 Day 20

Intestinal Crypts- Cancer Case

Can we predict the behaviour of irrational, unpredictable people?

Human Collaboration

1 1 0.995 0.995 0.99 0.99 0.985 0.985 0.98 0.98 0.975 0.975 0.97 0.97 E I E I E I E I S N S N S N S N 0% 5% * 10% * 20% * 0% * 5% * 10% 20% * 1 1 0.998 0.998 0.996 0.996 0.994 0.994 0.992 0.992 0.99 0.99 T F T F T F T F J P J P J P J P 0% * 5% * 10% 20% 0% 5% 10% 20%

Can we make predictions on graph- structured data?

• Identification of role, dependence, and influence in social networks. • Predicting outcomes of corporate merges, social acceptance, impact of economic legislation, and (of course) who your next social media friend should be!

• Enriching isolated models with appropriate background from knowledge graphs, i.e. bringing ‘common sense’ to machine learning.

• Predicting chemical properties such as the effectiveness of a drug against a given condition, as well as prediction of new chemicals given a set of desired properties. • Predicting structural information for amino acids, proteins, and enzymes.

• Fraud analysis/identification • Network analysis and cyber security • Logistics optimisations • Training data enrichment

Smart Connectivity

AI will always be inspired by natural systems What did brains evolve to do? Predict the unpredictable.

The future of AI Our next challenge is moving beyond simple Machine Learning We shall predict the behaviour of complex systems. To achieve this, we use Machine Learning that resembles the brain.