Featured Stories

No, Google Duplex Hasn’t Passed the Turing Test

We’ve just relaxed our human ambition

Junaid Mubeen
May 13, 2018 · 4 min read
Photo by Ethan Miller/Getty
Psst...there’s an audio version of this story. Upgrade to listen.

Am I the only one who feels a tad underwhelmed by Google Duplex? My social media feeds are filled with gushing platitudes over the virtual assistant’s handling of a simple haircut appointment. Some are even claiming the Turing Test has been passed. Putting aside the ethical ramifications of abdicating such tasks to the robots (a topic worthy of its own post), I’m left wondering: is this how low we’ve set the bar for human conversation?

Google Duplex may score top marks for authentic tone and delightful mannerisms, but the topic of conversation was hardly riveting. The virtual assistant negotiated a hair appointment through rote responses with dialogue that left no room for spontaneity.

I might imagine a small handful of calls that I would abdicate to a robotised assistant. I gain no pleasure from ordering prescriptions or querying bills over the phone. But there are some conversations I would never compromise on, such as setting up my next visit to the local board-game cafe. The banter that often ensues when searching for a slot amid busy schedules is hard to describe, let alone emulate. Needless to say, my beloved nickname at the café (‘Spiced Chai’, if you must know) was birthed during one of these calls. I will opt for this human brand of dialogue, and all the unexpected quirks it brings, over incremental time savings.

The fawning over technological feats might say more about how readily we lower the bar for human potential. We see a virtual assistant capture one tiny facet of our human behaviour and leap to the conclusion that there is nothing left for us. If the goal of the Turing Test is to emulate human behaviour, we may pass it simply by diluting those very behaviours.

I suggest we critically reflect on these newfound capabilities of our favourite toys by pondering the distinctly human capabilities that, at least for now, are beyond the purview of these innovations.

Google should devote as much energy to showing us what its virtual assistant can not yet do. It would have been more instructive for Sundar Pichai to put Duplex through its paces with rich conversational prompts. What better way to shine a light on subtleties of human conversation that are not so straightforward to replicate?

If the purpose of these technologies is to amplify our human potential, we may start by plumbing the depths of what we humans are capable of.

Case Study: Why virtual tutors will not replace teachers

In EdTech, my own line of work, many believe that artificially intelligent tutors are poised to replace teachers. AI is steeped in a narrative of displacement, one that conveniently slots into the teacher retention crisis. Why front the labour costs of human instructors when automated versions will do?

The answer, of course, is that they won’t. For all the impressive capabilities of intelligent tutors (they are certainly impressive enough to justify being called ‘intelligent’), these technologies are premised on knowledge transmission. Want to teach a student a body of core facts and procedures, in a sequence that cleverly adapts to their retention levels? Want to do this for millions of students at a time? Want a stash of real-time learning analytics at no extra cost? Say no more — the intelligent tutors are at your service.

But if you want deeper learning experiences, grounded in empathy and compassion — the most human, and therefore the most important, facets of teaching (more important even than content and pedagogy) — then please, for the sake of your students, keep human teachers in the frame. Abdicating these core tenets of teaching to a virtual being is the surest way to curtail your (and your students’) ambitions.

But don’t throw out the virtual tutors altogether. There is a twisted irony to the mantras of hardened traditionalists in education. They will reject most tech novelties, scoffing at the suggestion that intelligent tutors can every play a role. At the same time, they insist that the sole purpose of teaching is knowledge transmission. With this hopelessly narrow definition of teaching, passing the Turing Test would be a cinch. You may as well let the robots come.

Better to reflect on all that teaching entails and separate the mundane aspects (like marking and admin) from those that are deeply human (like authentic dialogue). As intelligent tutors automate the drudgery of knowledge transmission, we will be forced to contemplate what’s left for teachers. And there is so much left: Just as I insist on retaining my human connection with the board game café, we must never surrender the most human parts of teaching to our silicon counterparts.

Let’s make sure that when an AI does pass the Turing Test, for teaching or any other endeavour, it’s a test worth passing — a test that accounts for the highest ambitions of our human potential.

Junaid Mubeen

Written by

Mathematics. Education. Innovation. Views my own.