AI today is making fewer mistakes, but that is perhaps its biggest mistake

mediumThis post was originally published by Himanshu Sharma at Medium [AI]

Persona Intelligence: Next leap of AI | Part 1

Image for post

Background

AI (Artificial Intelligence) as a concept has been with humans for more than 70 years and it is one of the favorite subjects for Sci-fi movies and fiction novels. Some examples are Skynet (Terminator series), Jarvis (Marvel universe) and The Machine (Person of Interest TV series).

AI is revolutionizing how humans interact with technology and we are making rapid technological advancements every year. Current AI development is more focused on assisting rational decisions, not actually creating Human-like intelligence, which has much more to it than just logical decision making.

Persona Intelligence is to develop an AI which not only makes deductions but also has a unique personality just like humans and can “Think”.

So, why do we need a machine with a persona?

Key reasons for our innovations and advancements as a species can be attributed to our imagination and non-rational thinking. So, if AI must help us in our evolution, it needs to learn how to ruminate and discover new ideas without the limitations of the human mind. Thus, Persona Intelligence will be critical for our next leap in Space Travel, Medicine and even human sustenance on Earth.

Before we move to the current state and challenges, let me provide you with some background about AI. Every generation has contributed to its steady progress, but we could not achieve a significant breakthrough till the early 2000s due to three main challenges:

  1. Data: Data became the bottleneck for two major reasons: a) Variations in real-life problems far exceeded our data collection strategies and b) the need for manual annotation. But this changed with the dawn of the internet. Now, we have a huge source of information where billions of people come online every day and add to this data ocean. Companies have also found innovative solutions to crowdsource data generation. We all deal with Captcha regularly to prove we are not robots, but behind the scene, the same information is used to train the robot (AI).
  2. Computation power: As data increased, we required much faster systems to process complex calculations. In this area, we have made significant progress over the decades. To put things in perspective, our 2020 smartphones have similar power as an early 2000s supercomputer. According to OpenAI, the demand for computing by AI has been doubling every 3.5 months since 2012, and to quench this thirst the next most promising technology is Quantum computers.
  3. Algorithms: Advance algorithms in any field are a function of the iterative test and learn process. The early 2010s saw tech giants like Google, Facebook, Microsoft along with some pioneer start-ups developing advanced AI classified as Deep learning (DL) and Reinforcement Learning (RL) algorithms. They could iterate an algorithm in a day that would have taken years to run earlier. The key focus of these algorithms has been solving problems related to text and computer vision (images & video).

Current state

In the last 8–10 years, we have seen a massive advancement in AI capabilities ranging from autonomous vehicles, language translators and gaming algorithms which can defeat World Champions. Currently, AI is driven towards making rational/logical decisions but what about non-rational decisions or preferences?

Let me explain this with an example:

Image for post
  1. Is this an apple or an orange?

AI can answer this question faster and more accurately than Humans now.

  1. Which one is better: an apple or an orange?

AI algorithms can go online, read people’s comments and understand which one they prefer and why. It can return an answer like: 67% of people prefer Apple and the key reasons are ease of eating, sweetness, and health benefits.

  1. Which one would “you” prefer between an apple or an orange?

AI has no answer to this question yet whereas humans can answer this question quite effortlessly which is driven by their personality and their answer may not be the same every time.

Challenges

Humans make about 35,000 decisions daily, out of which the majority are not logical decisions. They are mostly driven by individual personalities. So, a true AI should have a unique personality (persona) as we all humans do.

Why have we not made substantial progress in this?

It boils down to 2 main reasons:

  1. We can only teach something which we understand thoroughly. Our rational decisions are easier to teach but teaching our personality /emotions is difficult, as we ourselves struggle with it every day. With the advancement in neuroscience, we might be able to decode our personality traits and then code them for a machine to learn.
  2. AI experts (companies & individuals) across the globe are obsessed with the accuracy rat race. Most of the new AI work is focused on error reduction. This process has created Sharp (attentive and accurate) algorithms but not Smart Algorithms with persona. Like most of the technological advancement in history, the need of the business is driving AI development. AI can now assist/replace people, reducing errors at an unprecedented scale. They ideally don’t want an AI with personality.

In conclusion, I would say, the next leap of AI must come from beyond DL and RL algorithms. AI should be allowed to make mistakes, not just to be more accurate but to make that experience part of its personality.

In the next part, I will discuss my point of view on the basic construct of a Persona Intelligence which has two main independent yet dependent components catering to rational and non-rational learning.

Spread the word

This post was originally published by Himanshu Sharma at Medium [AI]

Related posts