I Can't Do That but I Can Do This

AlphaGo's triumph over Grandmaster Lee Se-Exercise feeds our fear that Artificial Intelligence will overtake humans in every walk of life. Perhaps they volition take our jobs or, worse still, decide we are surplus to requirements and kill us off. I don't think we should be and then worried. Here are some things an AI tin can't do.

1. Answer The Ultimate Questionof life the Universe and everything

Alan Turing discovered the starting time limitation on Bogus Intelligence; it can't answer everything. Mode back in the 1930s he solved a famous mathematical puzzle chosen the Entscheidungsproblem. The puzzle asks if at that place is a universal problem solver that can solve any question you lot throw at it. Turing invented a Universal Machine – we now call this a reckoner – then asked what would happen if it ran every possible program, would that reply every trouble? The respond is no. Amazingly, running every possible estimator programme – even an infinite number of them – does not solve every possible problem. For the mathematically inclined amongst yous, this is considering there are different types of infinity. Computers work with the countable type while mapping issues to solutions are part of a large infinity, called the continuum.

BBC Hitchhikers Guide to the Galaxy, nonetheless from youtube.com

2. Solve Annoying Interview Puzzles

Maybe you don't want your AI to solve everything, just do some useful things such as calculate your expenses and play chess. A Russian mathematician chosen Yuri Matiyasevich used Turing'due south theory to show computers can't even solve simple interview puzzles. Yous've probably encountered these: Mary has some marbles and gives half to Sam and a third to Angela, how many marbles did Mary accept if… Solving these types of problems with an algorithm is known as Hilbert'south tenth problem and in 1982 Matiyasevich testify there is no general way to solve these sorts of problem.

world wide web.shutterstock.com zetwe

3. Write Bug-Complimentary Software

Another area that is hard for Artificial Intelligences is finding bugs in software. Don't get me wrong, artificial intelligence may be a big help in looking for patterns that might advise there is a bug only a piffling-known theory by Gordon Rice extends Turing's original proof to bear witness computer programs can't know annihilation interesting about some other estimator program. So an AI can't know if a estimator program is bug-complimentary, whether it will blue screen or whether information technology volition give y'all an out of memory error. Indeed, i estimator programme cannot tell yous anything not-trivial near another and since AIs are, at their heart, nothing more than computer programs they tin't assist you write issues-gratuitous software.

www.shutterstock.com Andrea Danti

4. Write software

The software we use in our daily lives contains bugs, just is all the same useful. Nosotros know computers can't find all bugs, but why can't they write useful software for united states? Fred Brooks explained inThe Mythical Human Month that writing software involves understanding the essential complexity of the real earth and turning that into rules.  John Searle's Chinese Room statement says AI does not have the homo kinesthesia of agreement. Bogus Intelligences can't write circuitous software considering they don't 'sympathise'.

www.shutterstock.com lassedesignen

5. Brand Moral Choices

Malcolm Gladwell points out in his volume Blink that people make instantaneous moral choices and and then analyze them after the fact to fit the rules. The choices announced to be emotional, non analytical. Artificial Intelligence only uses rules. A theory by Kurt Gödel says that any time you have a system with formal rules at that place will exist bug those rules can't respond –  the system will be incomplete. So although computers can obey the law –  and we had better bustle up and program them all with Isaac Asimov'due south three laws of robotics – they tin can't deal with new situations.

 Moral tests from wikipedia

6. Predict the Future of our Universe

There's a growing view in the scientific community that our Universe might exist a computer and we live in a simulation. We shoot probes out into space and simple equations predict their path. Peradventure this is considering the Universe is an enormous computer and uses these same equations. Only nosotros can test this using the Bell experiment. When we exam our Universe to see if it follows a pre-defined set of rules we find it doesn't. The particles in our Universe don't know what they will do in accelerate - there are no hidden variables controlling them - they behave randomly. Computers follow a strict programme with no randomness allowed so an AI tin't model our Universe and predict its future.

7. Win the Imitation Game

Alan Turing asked how we could tell a machine was intelligent given there is no definition of intelligence. His proposal was to put the supposedly intelligent motorcar in a locked room and enquire it questions. If the machine could fool y'all into thinking it was human being, then it was effectively intelligent. This is known as the Turing test. There'southward an easy way to crush the AI, ask it one of the incommunicable interview questions I mentioned before. We know computers can't solve these. Just there'due south a trouble with this thought. Mayhap humans can't solve these questions either! Are there any examples of humans solving impossible questions? The reply is yes.  In 1986, Andrew Wiles solved Fermat's Last Theorem, a problem a calculator cannot solve. Information technology took him nearly 40 years, just a computer could never have solved it. Humans will e'er beat an AI in the imitation game if you requite them long enough.

© James Tagg

8. Perform Jazz

Bogus Intelligence can't etch music that does not follow rules. Although jazz involves the unwritten exchange of rules between musicians every bit they play, musicians tin can evolve those rules and break them at will. Computers can't pause the rules and then their music volition always be formulaic. Roger Penrose has demonstrated non-computable mathematical puzzles and I am proposing music is a similar not-computable puzzle. Humans solve information technology when they brand music. It is quite hard to tell the difference between human being and calculator music, but there is always a difference.

Wikipedia

9. Invent
Humans invent things, compose music and prove mathematical theorems. Roger Penrose & J.R Lucas argue Artificial Intelligence tin can only follow the rules. Humans, on the other hand, step out of the box and create genuine innovation.

Photo by Doug Kline

10. Exercise Free Will

AI based on digital computers must follow the rules of its program. Those rules are deterministic and therefore don't allow for whatsoever free will. A theorem proposed by Conway and Kochen suggests the key particles in our Universe exercise accept gratuitous will and humans tap into this underlying free will then that their actions are non predetermined. Volition we be able to use this fact to requite our machines free will? I believe we will, but when nosotros do so, they will no longer be AI they will be Real I.

Flickr, Sarah Fagg

"Artificial Intelligence is similar an bogus plant. Information technology gives many of the same benefits but it's not the real matter."
James Tagg

sollarspontme1937.blogspot.com

Source: https://www.linkedin.com/pulse/10-things-artificial-intelligence-cant-do-james-tagg