AI 'likely to be more disruptive' than first five years of social media

Friday, 3 February 2023

Artificial intelligence is on the cusp of maturing, and is likely to prove more disruptive than social media or search engines were in their first five years of existence, Auckland University professor Michael Witbrock says.

Witbrock has a PhD in computer science, once led the Learning to Reason group at IBM and has his own brain-computer interface start-up called TransAxon.

As AI chatbots emerge that feel like speaking to real people, Witbrock says many people may be underestimating its likely impact.

Imagine “Google on steroids” – an AI-powered search engine that can continually learn to tailor answers to provide exactly what the user wants to know, and interact with the user like it’s a real person.

**READ MORE:

* I’m trying to make myself feel better after my interview with a robot

* Back to school: How will we stop students cheating with AI technology?

* What is ChatGPT and should you be using it for marketing?

**

Or a QAnon-esque actor that does not just push out misinformation or spread conspiracies, but can conduct thousands of one-on-one conversations with impressionable listeners simultaneously, tailoring responses to each, and continually refining methods of influencing them.

OpenAI’s ChatGPT bot has revealed how closely chatbots can mimic a conversation with a real person. The technology is continually being improved and refined, Witbrock says.

AI as a method of influencing

There have been plenty of recent examples of social media being used to influence – suspected Russian bots' involvement in the election of Donald Trump and Cambridge Analytica’s involvement in influencing the Brexit result are two recent examples.

“Whenever humans do this sort of influencing, they either do it one-on-one or to many people at once, but what we cannot do is have a conversation at the same time with a large number of people and pay attention to each,” Witbrock says.

AIs could take it one step further – they could have all these conversations while profiling the listener, and identify whether that person might be useful to its aims.

“It can really learn what the mood of the crowd is, and what groups are within the crowd, and which bits can be put together, and these are the sorts of things these systems are pretty likely to be able to do, and that’s a worry.”

That desired outcome could be voting a certain way, or it could be something more immediately dangerous, like rioting or attacks.

Unlike the dangers of influence via social media, Witbrock says we cannot afford to be behind the 8-ball on AI.

“I don’t think there’s anything proactive we can do about this apart from monitoring for it, both at a policing level and at a media research level,” he says.

“It might be good if research on this possibility were partly done in New Zealand, perhaps in part as part of our society's response to the Christchurch Call, and perhaps in cooperation with tech and media companies.”

But while there are risks with AIs, Witbrock says there are also potentially equal large benefits, and we should not limit the potential of artificial intelligence just because of the potential dangers.

“What most people see is some problem, and say ‘we should ban all this stuff’ without realising this thing that’s a problem is also an opportunity to do something good, that we cannot do now because of our limitations.”

Take a chatbot’s ability to speak to thousands of people at once. That could be incredibly useful if used ethically, for example in a state of emergency like Auckland’s recent floods.

Chatbot conversations with residents could ascertain not only whose driveways might be blocked by debris, but which should be cleared as a priority, for example if the occupants were in need of medical attention, or they had critical jobs, like nurses.

Automated checkouts that actually work

Automated checkouts portray Witbrock’s point of benefits versus risks.

Right now, Witbrock's opinion of automated checkouts is not high.

“They drive me nuts. I know how to use machines, and they fail and require someone to be summoned almost every time,” he said.

He says they could be vastly improved with AI and the likes of face recognition tech, which might stop the requirement for staff to check IDs when shoppers were buying alcohol.

But they have to be designed in a way that doesn’t threaten privacy.

But many want facial recognition tech to be banned, even though it could be created so it doesn’t store images of the face, and only uses cues to estimate the age of the buyer.

“You can have safe age estimation.

“This knee-jerk reaction of ‘this thing could be bad, therefore let’s ban it and keep on doing things badly forever’ doesn’t seem to be the way to go.”

Google on steroids

Witbrock says AIs aren’t far off being able to use short term, long term, and integrated memory abilities to look things up – similar to what Google does, but far better.

“It doesn’t just look things up, but it kind of works out what it should remember from what it’s worked out so far.”

He says AIs do not have these abilities yet, but giving them the ability is not a fundamentally hard thing to do.

He says using chatbots like ChatGPT to search will be unreliable in the early stages, but it will not be long before they become more reliable than most people, because they can draw on far more information and processing power, and are able to continually refine responses to what a questioner needs.

The first outfit to produce a Google-beating AI could supplant the tech giant as the new gatekeeper of the internet, as it wrestles dominance in online search (and the advertising that followes).

While OpenAI might be a first-mover, Witbrock says the public should not make the mistake of thinking they are the furthest ahead.

Google and Facebook are “top-tier” when it came to AI development. One engineer believed Google’s LaMDA system was now sentient.

Witbrock says indicators in the industry suggests Google is not far off deploying something quite soon that might help secure its dominance in a new AI-driven search engine race.

“OpenAI did a kind of brave, or reckless, thing, in making this (ChatGPT) available because it’s really focused our attention on it, but there are a lot of players in this game.

“Facebook has really good researchers in this area. Baidu has really good researchers in this area.”

Current limitation to chatbots

Witbrock uses an analogy to explain where chatbots are now, compared to where they could be.

Right now, they operate like a human does when the stakes are low, and loose answers are acceptable.

”When you ask your crazy uncle to tell you what happened on the news yesterday you don’t expect very much from it – you don’t expect to rely on what they say.

”When you ask your best friend for advice on a medical problem, again you don’t expect too much from that.”

The current chatbots are at that level – they will have a quick crack at an answer, but they don’t have the expertise of, or aim of being, an expert.

Humans change their methodology when the stakes are higher. They use more professional language, they follow a stricter process, and they reference proven facts.

AIs have to undergo the same change, and once it does, its usefulness goes exponential.

For example, it could assist doctors, having read every piece of literature on a given disease.

Witbrock says he moved back to New Zealand from the United States, because the country has the potential to lead the world in rolling out AI into society safely and rapidly.

“We can be a model for the rest of the world, and make things better here because of the sort of society we have using AI, and I’d hate to see us miss that change.”