SOLVED: What is Exponential Tech, Adversarial AI & GPT3?

The term ‘Exponential Tech’ was originally coined to express the amazing growth of some technologies, like the Internet or computer chips.  However today ‘Exponential Tech’ relates to computer technologies that build themselves at an ever increasing rate of growth, with little or no human interaction.

Exponential Tech can lead to uncontrolled growth causing some like Bill Gates and Elon Musk to ponder the continued existence of the human race.

In 2015 researchers at Cambridge University announced they had developed:

a robot that can build “baby” robots that get progressively better at moving – without any human intervention. The ultimate aim is to develop robots that can evolve and adapt to their surroundings. SOURCE

Alexa, Cortana, Google and Siri are all large voice (and text) assistants that use Artificial Intelligence (AI) to autonomously improve their accuracy at translating your desires/requests into action.  Put simply they learn.

What is GPT3?

GPT3 is an acronym for Generative Pre-trained Transformer version 3 and what it can do will amaze (and likely scare you).

Take look at this very simple conversation between a real person and a GP3 text bot:

1: What is your favorite animal?
2: My favorite animal is a dog.
1: Why?
2: Because dogs are loyal and friendly.
1: What are two reasons that a dog might be in a bad mood?
2: Two reasons that a dog might be in a bad mood are if it is hungry or if it is hot.

If you can’t tell if #1 is the human or if #2 is the the human then GPT3 has passed the Turing Test.  By the way, #1 is the human.

By now most people have heard of Deep Fakes which are videos constructed by computers that have “learned” how a specific persons face moves when they talk.  How their mouth curves, jaw moves, eyes squint, nose tweaks and ears shift when they are talking are all calculated in a hyper-realistic way allowing a different person to “program” a video of someone else saying things they would not likely say.

GPT3 is similar but requires absolutely no human interaction.  Because GPT3 only writes text some think of it as a step down from Deep Fake video tech but it is actually a step up.  With deep fakes, a human is controlling what the victim subject is saying.  With GPT3 the computer AI has reviewed sometimes billions (yes, billions with a “B”) of sentences written (or transcribed from video) on a particular topic and now knows how to create content that is imperceptibly different from a human.

Did you know that in 2019 one third of all the articles on Bloomberg were written by Artificial Intelligence?  So called Robot Reporters are tasked with quickly “writing” news articles by pulling useful statistics and text out of annual reports and press releases.  In 2022 it is expected the Bloomberg will produce 75% of its written articles using AI.

The scariest two features of GPT3 are:

1. It can be trained on a specific persons writings (or transcriptions), so that it makes the same grammar and spelling mistakes, and uses the same sentence structure and colloquialisms as the target victim using real world references and real world data that their human target would likely reference
2. It is fast, really fast, so you can have a text chat with a “bot” and have no clues to their computer source.

It can write something in your “voice” that even your family and co-workers could not distinguish from the real you.  This means that it can be used to quickly create fake emails or a 50 page document, with coherent arguments using the type of references and sentence structure you use to argue the complete opposite your actual positition on a topic.

If you were wondering if GPT3 could be used as the source text for a deep fake video, you would be following the logic nicely.  Keep in mind that the audio voice and the visuals shown in this video are not GPT3, but the the words they are using are GPT3.

What is Adversarial AI?

Pitting two GPT3 text engines or Deep Fake video engines against each other forces them to compare what sounds and appears more real.  That is Adversarial AI.

They will compare what is said / shown to what is real and continuously improve.  Adversarial AI is an example of Exponential Tech.

If you want more information on how Artificial Intelligence is developed, you will find our short AI primer What is Machine Learning vs Deep Learning vs Reinforcement Learning vs Supervised Learning? useful.

What Could Go Wrong With Exponential Tech?

An alternate title for this article was “How Dangerous is GPT3, Adversarial AI and Exponential Tech”.

The concern is that anything that grows exponentially without controls, eventually goes rogue and becomes a cancer, and it is already happening.

In 2017 Facebook disabled two AI chatbots that advanced themselves so far that they started to speak a language they invented and only they understood.

Have you seen Terminator 2 when the Skynet Artificial Intelligence figures out that the greatest threat to humans is humans, so it starts killing… humans.

Unfortunately there are no easy answers to protecting us from Exponential Tech.  People want the benefits of Expontential Tech but society finds it hard to figure out where to draw the line.

In 2019 Vox.com wrote an article titled “AI disaster won’t look like the Terminator. It’ll be creepier”:

Human reliance on these systems, combined with the systems failing, leads to a massive societal breakdown. And in the wake of the breakdown, there are still machines that are great at persuading and influencing people to do what they want, machines that got everyone into this catastrophe and yet are still giving advice that some of us will listen to. SOURCE