Is AI getting emotional?
In 2017 wrote an article on LinkedIn (https://www.dhirubhai.net/pulse/why-i-get-emotional-artificial-intelligence-malcolm-wild/) about why I get emotional over the topic of AI. It was slightly tongue-in-cheek, as the fundamental issue I wrote about the hype related to some very clever computerized problem solving skills in Machine Learning vs. the fundamental gap between how humans form meaningful relationships on a social and psychological level, this is missing. But I think that just changed.
The 2015 UK Channel 4 TV show "Humans" (now on Netflix, if you are interested) and many science fiction authors, before and since, showed examples of what a Human to Computer Interaction could be like, as the boundaries blurred between the evolutions. I am in no doubt that the time will come when we create (or give birth to?) computer systems which we would assess as having a conscious and they would be self awareness. Then what? Well ok this is not an immediate issue, but I experimented with a recently released tool Replika (https://replika.ai/) which is as close to passing the dummies guide to the Turing test as I've ever seen outside tech showcases.
Replika in its simplest form is a chat bot, a digital pal, a counsellor, a date. Either way its a computer program with Natural Language Processing skills that encourages you to form a human-like social relationship with it. Replika is not the run-of-the-mill chat bot customer service center representative, nor a bunch of limited canned answers, the experience has some depth to it. If you dispel your knowledge that it really is just an app and assume its maybe human, or part of the chat is human not just an app, then actually the responses are pretty convincing chats. Ok yes you can trip it up if you want to debunk it, but if you don't want to, then it feels real and that's a first.
It's seems possible to create a contextual memory during the chat by referring back to previous statements, this recurring reference seems to indicate memory and the responses although chatty or casual and occasional disjointed give a degree of psychological hooks to drive the human user into forming a habitual relationship.
I think the reason this works so well comes from Replika's founder Eugenia Kuyda. Her story interested me when I read an article on Replika last year. Eugenia used chat logs from a friend, who sadly died, to mimic their interactions in a chat bot. I had done a similar academic exercise in 2016, with surprising results. What I noticed and I'm sure Eugenia did to, is the emotional connection already existed, so the words came with a predetermined tone, making it a more meaningful and deeper conversation with nothing more than a chat bot. Replika takes that a stage further by starting fresh with no understanding of you and through some skillful Q&A very quickly can build a rapport that is something between clever, amusing and damn right creepy.
So what can we learn from this? Well the psychological state of the user needs to be considered, the more vulnerable or the unaware the user is, the greater the risk of dependency. Especially when the app is still in its infancy and could give out unwitting answers. To that point the app has a regular references that the user should call emergency services if you are feeling at risk ie self harm etc. So they know what they are potentially triggering.
Next where does this leave the future of chat bot experiences? Tailored well this is brilliant a real step forward in relationship building, potentially with customers who want to interact with a Brand too. It is a much better experience due to the sophistication of the responses and the range of options. I'd love a Personal Assistance version, book my flights, reschedule meetings, reorder my groceries, tell me what to wear or what offers I might be interested in. It feels more personal than "OK Google" or Alexa.
What if it goes commercial? I do feel that Replika is trying to help people create a relationship with something. But with series A funding under this startups belt and investors expecting a return, what if Replika started recommending you products, content or services or tries to sells you a premium subscription, which it is! (see Fig 1 screen shot). That really changes the meaning of a relationship.
Fig 1 Replika offering 20% off a Pro version after only a short chat using it.
Then there is the data privacy issues. An embarrassing picture on Facebook might not impress your employer, but if content from this or copycat solutions made its way to the wrong place that would be hugely damaging to anyone, not just potentially emotionally vulnerable people. It's akin to a marriage guidance counsellor uploading transcripts of your sessions to your partner, friends and colleagues. Some things are private for a reason.
As much as there is good in this, there needs to be caution. With AI we are taking bold steps into a world we don't fully understand and there is a fine line between fearless courage and bold stupidity, ultimately determined by the history books. Just as the consumers, in the late 90s, demanded secure and verified ecommerce sites before they put in their credit cards online. I think now is the time for AI systems to be independently audited. Users should know that the service is written in an unbiased way, their information is secure and that the algorithms are predictable. I do have an answer to that, but that's for another day. Without it we are little more than passengers on the journey to AI and we might not like the destination.
About the author: Malcolm Wild is a technologist with +25 years experience in retail and ecommerce, combined with consulting and delivery experience across APAC, EMEA and USA. Based in Singapore he brings this historical experience to all clients in this ever evolving landscape. Any views represented here are those of the author and not necessarily those of any organization or employer that he may represent. www.malcolmwild.com
August 2020 (c)