The Singapore Institute of Arbitrators invited me to debate the following motion: ‘This House Believes That Artificial Intelligence Will Have Replaced Arbitrators within Twenty-Five Years’. In short: can – will – algorithms replace arbitrators within a generation?
We were debating this last night. Here are my opening and closing statements.
There were two debating teams and a tribunal of three judges, Alessa, Jae Hee and Ankit, all peers. Sapna took over moderating at short notice. This meant a good time was guaranteed. Katie and Divyesh argued for the motion, Benson and I argued against it. In the end, we ‘won’ the debate by the judges’ decision and the audience’s show of hands, but compared to a vote at the beginning, the proponents of the motion persuaded many in the audience to change their minds. ‘A moral victory on both sides’, as Andy from the SIArb put it (his account of the evening is here, by the way).
I guess what I’m trying to say is it was great fun.
The Opening Statement
Hallo everybody. Thank you, Divyesh, for your remarks on why you believe artificial intelligence will have replaced arbitrators within twenty-five years.
That won’t happen. And why not?
For the same reason why a robot vacuum cleaner isn’t, and won’t be within the next twenty-five years, good at football.
Why do I say that? Don’t a robot vacuum cleaner and a human football player have a lot in common? For example, don’t they both move in a field which is staked out lengthwise and widthwise? As such, isn’t it conceivable that we develop robot vacuums which can cover the dimensions of a football pitch? Isn’t it conceivable, then, that we develop robot vacuums which do so in line with the rules of football but much faster, more enduringly and perhaps even with better overview than any human football player?
Yes. But football isn’t a two-dimensional game.
In football, goalkeepers dive for saves and field players send over crosses. The ball is thrown in from the side line and kicked through the air where it’s often headed. Sometimes, rarely, you see a player perform an overhead kick.
All these actions happen in the third dimension of height. Football wouldn’t be the same without them. Rather, it’s fair to say these actions constitute football. But they are literally above any robot vacuum cleaner that may stray onto the pitch.
AI and human arbitrators have a lot in common, too. In principle, both are capable of establishing the facts of a dispute and the rules applicable to it, and of applying one to the other. Nowadays the difference between them doing this is merely gradual. Thus, isn’t it conceivable that within twenty-five years we develop AI which does these things much faster, cheaper and less error-prone than any human arbitrator?
Maybe. We’re already well on the way. There are self-learning algorithms and all. But it would be wrong to confuse a picture detail with the big picture.
The game of arbitration isn’t two-dimensional either. In addition to establishing a set of facts and applying a set of rules to them, there’s the handling of the parties and their counsel throughout the proceedings and certainly during the hearing. There’s the reading between the lines, the knowing or at least sensing the existence of cultural idiosyncrasies. The capability to innovate, not just to emulate. To grasp context. Be empathic. Have a sense of irony, where applicable.
All these capabilities and skills happen in the dimension of consciousness. Consciousness is the ability to feel things: joy, pain, love, anger. Arbitration would not be the same without it. Rather, it’s fair to say conscious actions make up arbitration. Conscious actions, however, are above any AI.
Three-Dimensional Robot Football
Of course we could set our minds on developing robots which play three-dimensional football. Even then, not many of us would have the cheek to say robots would replace football players within twenty-five years. It’s just too challenging to bring the locomotor system of robots on par with ours. But we could develop these robots and improve them continuously if we wanted to, and why? Because in addition to length and width we fathom the third dimension of height.
Three-Dimensional AI Arbitration?
On the other hand, we seem to be incapable of creating artificial consciousness. That’s because we don’t get consciousness ourselves. The understanding of consciousness is not only above AI, it’s also above us, remarkably.
For decades we have been conceptualising and for many years we have been developing AI. We are improving it now, at dizzying speeds and with fascinating results. But in doing so we haven’t elevated AI one iota into the sphere of consciousness.
Nevertheless, AI beats us at Jeopardy, chess and Go now. It’s really smart. Which is why arbitrators ought to use AI and ought to use it more as it develops – don’t get me wrong on this!
AI has become so smart, if we had put the same effort into developing football-playing robots, they would be amazingly fast in going after the ball, unrivalled perhaps at game tactics and brilliant in anticipating offside positions. We might have seen the reinvention of low-pass play. But not many matches would have been decided by such a robot.
It seems reasonable to continue improving AI as tools, as tools for arbitrators and other dispute resolvers as well. After all, AI is good at sorting and structuring dispute-relevant data, and fast, and it can help us predict the probability of case outcomes.
But if we indulge in the fascination for AI and replace arbitrators with algorithms, the result will be a quicker, here and there even improved, and yet poorer form of arbitration. Without sense for nuances or peculiarities, without grasp of context. Without capability to innovate, only with the ability to emulate. Lots of IQ, but no EQ.
Like football played in two dimensions only.
That’s why this won’t happen within twenty-five years, neither in football nor in arbitration.
The Closing Statement
Dispute resolution by AI seems to be a threshold issue in that it will require of algorithms something which, as of today, they don’t seem to master.
Nowadays AI can be a tool.
To become a dispute resolver by itself, AI would have to be capable of comprehending both the facts of a dispute and the rules applicable to it completely, and of applying one to the other. This would be the easier (I’m not saying: easy), the easier describable the facts and rules applicable were.
But often they’re not. There are fuzzy things like ‘appropriate interests’ or ‘context’.
An example, by English cognitive scientist Margaret Boden: a man and a woman go out into the forest to pick blueberries. He’s twenty years old and can pick ten pounds of berries per hour, she’s eighteen and picks eight pounds an hour. How many pounds will they bring home an hour later?
The answer of course: maybe zero.
Try to get that answer out of an algorithm!
There are vague legal concepts, too. And there isn’t precedent for everything. If there’s a dispute to resolve but no precedent to be found, what to do?
We Get the One, We Don’t Get the Other
Judges and arbitrators of today deal with these imponderables by applying (i) their intelligence and (ii) their consciousness.
We’ve made a lot of headway in artificial intelligence. We haven’t made similar progress in artificial consciousness. Of course, perhaps consciousness as an element of resolving issues could be substituted with something else. But we are even more clueless about what that could be.
But why isn’t it enough to resolve a dispute using intelligence? Why the need for consciousness? We submit, because humans like their dispute resolution not to be correct and just alone – whatever this may mean –, but to be reasoned also.
Reason, however, is the capacity for consciously making sense of things, establishing fact and verifying facts, applying logic, and changing or justifying practices, institutions, and beliefs based on existing or new information. In short, the capacity to apply to a given situation everything a grown-up human has learned in her life.
Compared with this, the task of, say, analysing a handful of non-disclosure agreements or keeping track of a few dozen Go stones is simple.
Another example, something I read in the Kluwer Mediation Blog the other day: all of us understand when someone says ‘a man walks into a bar’. We all know this is the format of a certain type of joke.
Also, all of us understand when someone says ‘a man walks into a Saudi consulate’. We are immediately reminded of the news context of only a few weeks ago. We all know this isn’t a joke.
But will an algorithm be aware of this context?
We assume parties who choose arbitration want their dispute resolver to be able to tell one from the other.
Different Ball Game
But sure, the day may come when a self-learning algorithm will be intelligent, albeit not conscious enough to resolve a dispute. If we tell it to do so, it will also disclose the steps of its decision-making. We might not be able to make sense of it, if it’s a neural algorithm which processes patterns associatively, but that’s a different issue. In any case, the algorithm won’t be able to reason it out.
But maybe, within twenty-five years from now or whenever, there will be parties, artificial or not, who put no value to reasoned decisions? Certainly AI could resolve disputes among these parties?
I’d hesitate to call this way of dispute resolution arbitration though. Just because it’s a ball game doesn’t mean it’s football.