Scientist Alan Turing introduced the Turing Test in 1950, a radical yet simple thought experiment to detect intelligent machines. According to the test, a technology that masquerades as a human convincingly can also ‘think.’ Six years later, the term ‘Artificial Intelligence’ was coined, charting a new course for human progress. Over 70 years later, much work is left to make sentient machines a reality. As an investor and an engineer, Aun Hussain has been involved in AI for over a decade. In his keynote, AI: Beyond the Hype and Buzz, he presented an overview of AI’s capabilities and the landmines we need to be wary of.
AI Through Several Lenses
So let’s jump back and ask: What is AI after all? What are its potentials, promises, and pipe dreams? It’s been dubbed slow to deliver, a sector where investors put their money and wait. But if you are to make an impact in AI as an angel or an entrepreneur, there are ways to look at these technologies.
Linguistically, AI has two components. The word artificial refers to the human-produced element of AI, for example, a robot or software. The next component is intelligence. An accepted standard definition cited by Aun goes something like this:
“Intelligence is the ability to solve problems or to create products that are valued within one or more cultural settings.” — Howard Gardner, psychologist
So, how do you measure it? As the Turing Test tells us, the closer AI can approach mimicking a human being, the more valuable and evolved it is. Let’s look at a few classifications from a low to a high level of evolution in what is called the Cognitive Pyramid.
Reactive Machines: Found at the bottom of the pyramid, these AIs react to inputs according to rules and have no broader context or memory/learning. Think chatbots and Amazon store robots.
Limited Memory: These occupy the next part of the pyramid’s lower half, referring to technology that uses observations to inform decisions in the future. Tesla’s self-driving cars fall into this category.
Theory of Mind: Aun says Open AI may head in this direction. What is this category? It’s a possibility that machines understand emotions, beliefs, and thoughts and can perform social interactions.
Self Aware: We’re in the sci-fi realm here. The apex is about self-aware, sentient AI machines, a future hidden for now.
Aun says that thinking in terms of Weak AI or Strong AI is also beneficial. It means asking about purpose and capabilities. A narrow, Weak AI solution is designed for specific tasks and does not possess general human-like intelligence, while a Strong AI solution can perform a wide range of tasks and understand/learn. Another way of looking at AI is a technology or domain-based view. Is the solution based on deep learning or image recognition? Is it a text-to-speech or an AI vision technology? AI is deployed for tons of other functions. Sometimes, this results in a mishmash of several technologies. In each case, one should know the purpose behind their deployment. We are beginning a new era, and the startups should rank well within the Weak/Strong map, Domain Expertise chart, and Cognitive Pyramid.
Small Truths on Large Language Models
Everyone is buzzing about ChatGPT. From an engineer’s stripped-down perspective, it contains a layer for inputs. This layer is built into the model, with the process culminating in an output. It would help if you told ChatGPT whether it is right or wrong so that it can change its output and come up with precise information. It is a Large Language Model that assigns probabilities to the next word in a sentence. However, LLMs only predict based on the input, and these models are not programmed to feel or learn.
So, what are the Strengths, Weaknesses, Opportunities and Threats of LLMs and AI?
Strengths: Unlike search engines, LLMs read the data sources and can be called “search on steroids”. It’s helpful for organizing data and creating content.
Weaknesses: LLMs are trained on data sets and, not on anything outside that data set. Well, what if the data is biased? AI can even include fake quotes by made-up people. Its models are generalized, there is no context deep meaning or nuances in perspective, it gives baseline answers that human judgment can and should make better.
Opportunities: Domain-specific expertise is about creating value from specialized data. This will define the future of AI, along with features like integration with other services. Can you increase productivity and use its data structuring qualities?
Threats: Quality of data is important too, as physical misuse in war, and societal misuse in deepfakes. It may even affect social cohesion and unrest when say Uber drivers or food deliveries inch closer to a future with less human involvement. This will be responded to by regulations, which are crucial for investors too.
So what are the regulations? In the US, the FDA, FTC, AND NTIA are introducing new rules for AI, and there’s talk of a Federal Law on the horizon. Italy temporarily banned ChatGPT, and in Japan, the Product Liability Act may hold the developer or operator of the Al system liable if damages arise from tangible objects (such as machinery or robotics). When you look at a startup you have to keep in mind these aspects and it helps to remember that as tech is evolving, public policy is slower to evolve. In a nutshell, looking at AI technologies from different perspectives and knowing specifics about the promise and perils goes a long way in making sure investors and founders ride the wave and move beyond the hype.
About the Speaker
Aun is deeply passionate about building products customers love. As a technical product leader, he is responsible for understanding “what” to build, “how” to build it, and “delivering” it to customers. He has a rare combination of technical, business, and entrepreneurial skills, as well as both startup and big company experience. At Amazon, he built a new Amazon subsidiary, worked on creating one of the world's most successful mobile money platforms at Telenor, managed innovation, P&L, and product lifecycle at Tyco (parent company of ADT), and led the messaging product portfolio at Bell mobility.
Watch Aun’s entire keynote here: https://keiretsuforum.tv/artifical-intelligence-beyond-the-hype-and-buzz/
