On which a computer passes the Turing Test for the first time
Yesterday, a computer simulating a 13-year-old boy named Eugene passed the Turing Test at an event organized by University of Reading at London’s Royal Society. This is a huge and remarkable breakthrough in the AI front.
The Turing Test is a test of a machine’s ability to exhibit intelligent behaviour equivalent to, or indistinguishable from, that of a human. It is, essentially, a conversation between three parties—Player A (human,) Player B (machine,) and Player C (human)—which are separated from one another. If the judge (Player A) cannot reliably tell the machine (more than 30% of the time) from the human, the machine is said to have passed the test. Until yesterday it had never been achieved. The Turing Test doesn’t check the correctness of the answers, rather how closely the answer resembles typical human ones.
Quoting Alan Turing:
I believe that in about fifty years’ time it will be possible, to programme [sic] computers, with a storage capacity of about 109, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning.
Yesterday also marked 60 years since Turing’s death.
This marks a very important step forward for the AI (and the general Computer Science) community. We’ve made—for the first time—a machine that is capable of being so smart and fluent with human language and cognition, one can’t tell it’s not a human. Such an event paves the road for superintelligent machines, better NLP, artificial neural networks, along with better answers to philosophical questions such as “can a machine have a mind, consciousness and mental states?,” “can a machine have emotions?,” and “can a machine be self-aware?”1
The rabbit hole goes deeper
The philosophy of artificial intelligence is a broad field and currently one of the most intellectually stimulating ones. Intelligent machines are closely related to questions about the deterministic nature of our universe; if it’s completely describable programmatically or even a simulation, if there are multiple (perhaps infinite; containing all logical possibilities) other universes (or universe simulations), if the simulation is written by another species or a sapient machine or if it’s a simulation within another simulation.
Some (like Price and Hamilton) argue that humans are self-replicating machines themselves. I briefly touched this topic at the end of my Google Glass review last year:
In 1967 George R. Price went to London after reading Hamilton’s little known papers about the selfish gene theory and discovering that he was already familiar with the equations; that they were the equations of computers. He was able to show that the equations explained murder, warfare, suicide, goodness, spite, since these behaviors could help the genes. John Von Neumann, after all, had invented self-reproducing machines, but Price was able to show that the self-reproducing machines were already in existence, that humans were the machines.
Juergen Schmidhuber famously said at his TEDx talk (which I can’t stress properly how much of a “must-see” it is; arguing about this universe and our own lives being just by-products of a very simple and fast program computing all logically possible universes) that “to the man with a computer, everything looks computable.”
Now, if our reality is indeed deterministic, then it’s also completely describable (and thus, predictable) by a computer. Hence, a superintelligent computer will have much more executive and cognitive power over this very domain. Moreover in this case it’s also easier to describe the world itself to this machine. Once I argued that, perhaps, “if our universe were to be a computer simulation then Deterministic Finite Automata would be to it what particles are to physics models” but my friend Panagiotis counter-argued my hypothesis with an even better, interesting, and in our case, extremely relevant to the Turing Test, one:
That’s very interesting topic and I keep thinking about it constantly.
Let me first give you a proof why particles are not the equivalent to DFAs. Your computer is the equivalent of a turing machine and it’s built by particles. So assuming particles are equivalent to DFAs, a model emerged from a DFA can’t be more powerful than a DFA. But, by definition Turing Machines are more powerful than DFAs, thus particles can’t be equivalent to DFAs.
What is more, a human beeing might be more like a turing machine. That comes from the fact that DNA/RNA itself seem to be a turing machine and models emerging from it can’t be more powerfull than a turing machine. Thus, a brain probably is equivalent to a turing machine, capable of running other turing machines as well.
So, here comes my point. Given that the brain is a turing machine, that means it can be simulated and for me that means that also the perception or the “soul” can be simulated as well. Free will is just a perception. For me and you, there will always be a machine that can simulate us. Laplace in a similar fashion introduced the thought experiment of a demon being able to predict the future given it has full knowledge of the current state. Determinism comes from the fact that someone knows the exact current state. Let me give you another example. For a computer there is no random thing. It can always know what’s the next number so for a computer there is only determinism. However, for the human perception, which might not have access to the current internal state of it, does this actually matter? Will you feel less lucky if you win the lottery from numbers generated by a computer?
I believe having full knowledge is impossible and thus determinism is well hidden under this constraint. In game theory that’s the equivalent of incomplete information. Thus, the free will is just ability of organisms to create strategies to cope with that incomplete information. Determinism doesn’t contradict free. It just emerges from our limited capacity of predicting the future.
Let’s not forget that a deterministic universe means that we lose our free will. As Panagiotis said above “Free will is just a perception.” Don Knuth said also something apt to me last year when I asked him about a related topic:
[…] and if our universe is a computer simulation, which means we’re simply mathematical representations and everything is deterministic and, as a result, we lose our free will, then there’s nothing we can do about it and we cannot answer it, thus we shouldn’t bother thinking about it.
It is obvious now that the Turing Test is relevant not only to abstract sapient-or-not machines for “conversation games” but also to physical and biological systems like the DNA. The implications of computability (and especially, intelligent computing) are enormous. Yesterday we went one step closer to intelligent computing; something we couldn’t fathom even a few years ago. What an exciting time to live in.
A new avant-garde field in bioethics is “theoretical bioethics” which argues whether software can suffer. And if so, what are the ethics of creating, modifying and deleting it from our hard drives. ↩
posted: June 9, 2014