Can Robots Be Effective Advocates?

Could we one day have humanoid robots pleading arguments in courtrooms, appearing on TV to persuade people of the political views of a newly formed “Robot Political Party,” or standing in storefronts convincing people to buy a new rug or set of steak knives? Put simply, could robots ever be better advocates than humans?

The idea of a robot advocate seemed tantalisingly possible after watching IBM Watson win the game show Jeopardy in 2011. With Watson physically standing between two humans contestants, listening to the presenter ask questions, searching through databases for an answer, and then speaking it in the form of a question, it posed the scenario of whether we could one day have someone wheel Watson into a courtroom and stand at the bench and listen to the judge’s questions search through and find answers and speak them back. Having built my own research AI in Ailira and used parts of the IBM Watson platform as well as others in chatbots and search engines, I can say that there is more to advocacy that just finding an answer. Some part of the advocacy process is finding answers to questions, but you need someone to ask intelligent questions. What a research tool like this does is leverage the abilities of the user. So Ailira enabled my girlfriend (now wife) Sarah, who is a speech pathologist with no knowledge of tax, to use only Ailira to pass the Adelaide University Tax Law exam. Ailira leveraged the abilities of Sarah to understand the questions and answers to tax problems, but in real life, as opposed to a game, there is required human skill in understanding what questions to ask. And a more skilled person can use the leverage of AI in researching to an even greater advantage because they can ask more precise and insightful questions and understand how best to answer them given the mechanisms of AI. This we move to a scenario of the superior advocate being a hybrid human-machine. Importantly the task of legal research is but a small part of advocacy. Anyone who has presented in court knows that they must present their case to the judge rather than merely answer questions that the judge asks.

GPT-3 (the Generative Pre-trained Transformer 3) is an auto-regressive language model that uses deep learning to produce human-like text. Created by OpenAI, GPT-3 can be trained on data sets and written based on that training. It has been used to write a published book of poetry, news articles, and academic essays. It could probably be trained on a database of court submissions and then produce them on demand. Much more than entering fields into a template, this would recreate an entirely unique text submission every time based on initial programming. But even if such a robot was trained on countless amounts of court transcripts and was able to produce submissions, it would still be inferior to a human as an advocate. Some reasons were explored in my earlier article ‘Why There’s No Such Thing as a Robot Lawyer and there Never Will Be”. Other reasons are as follows.

Advocacy is more than presenting a set of facts. It is constructing them in such a way so as to best appeal to the audience. Humans are social creatures. Like it or not, we are influenced by those that we are around. A skilled advocate can use their charisma, empathy with the listener, and story-telling narrative skills to move someone onto their side. While there are AI that can manipulate people based on big data, as we have seen with, say, Facebook and YouTube algorithms, this is more about reflections and presenting people with more of what they have expressed that they like rather than convincing them to like something that they might not otherwise. Advertising is not advocacy.

While we might have robots involved in elections, infamously re-tweeting comments, and disbursing fake content in recent elections, humans are enamoured with stories. Whether it is a rise and fall and rise again of a hero or a battle between good and evil, ultimately, the lead character in our stories has to be something that we can associate with. I could never see people associating with an eternal robot that has a ‘junk to jewellery’ story of its rise. In particular, because the story might have no end and could last eternally – stories must have an end. Mortality ends all human stories eventually, and that promise enables us to back the hero. An immortal politician seems to me to only be a villain to be conquered. I cannot see how humans could be persuaded by a robot politician advocating their position. It might be possible for scientism or some worship of science to lead to some god-like robot entity or even a fictional persona. But I think that there would need to be humans advocating on its behalf.  

While the best-used car salesman might always be a human who can size up the potential customer and sell to them based on particular desires and weaknesses, it is well known that salesmen will often use repeated sales scripts. So I am certainly not saying that there will be no robot salesmen. There have been books on how to write sales letters for a hundred years. Online sales, conducted entirely without humans, have been increasing exponentially. However, when trying to sell a particular product to a particular person, a talented human always has the potential to do better than a robot.

The best use of AI in law is to leverage the existing capabilities of humans. While this might mean lawyers doing faster research or writing faster submissions, this would also mean the unskilled using AI to leverage their abilities. There are huge problems with access to justice. There is a potential to ameliorate this by having unrepresented or partially represented parties use AI to assist them to answer questions so that they can better understand their facts and circumstances, to take down their circumstances and arguments so as to order them in a manner to be better placed before the court and to assist them drafting scripts and submissions that they could make. While, of course, it might be inferior to the best advocate, such robots would be better than the worst.