11.1.2022 Executive Data Bytes – Unthink AI - Time to go outside of the box

Executive Data Bytes

Tech analysis for the busy executive.

Welcome to another edition of Executive Data Bytes! This week, we are talking about why you should be thinking differently about AI. 

Focus piece: “Why we need to think differently about AI”

Executive Summary

While AI researchers are racking up impressive successes, it isn't surprising that AI isn't closer to solving general intelligence or creativity problems. Learn about the shortcomings of AI and how we can improve on these in this O’Reilly article.

Key Takeaways

• Ellen Ullman's critique of abstraction in Life in Code is a good way to think about how artificial intelligence (AI) has permeated our notion of artificial intelligence. AI systems often give incorrect answers, because they don't know how to say "I don't know".
• David Chapman suggests that one problem with AI is that we don't know how to evaluate progress. We don't have the abstractions to understand intelligence, so we just watch the demos, which isn't good enough to understand what AI can do.
• AI is an abstraction engine. If our abstractions are the problem, then second-order abstractions to measure our progress are not going to point us in the right direction.
• The phrase "Big data is people" has gained currency in the discussion of data ethics, and it's important because it's a warning against abstraction. Data abstractions are often the basis for decisions, including loans and mortgages, bail and prison sentences.
• When we think about a future AI, should we be thinking about the human mind's ability to form abstractions, forget, form models with minimal data, change our minds, get bored, find things "interesting" or "beautiful" for no reason other than that they please us, and make mistakes?

Focus piece: “AI and humans think differently”

Executive Summary

Humans learn by making mental concepts that link together many different properties and associations. AI systems rely entirely on extracting complex statistical associations from their training data and then applying these to similar contexts. This article from The Kathmandu Post explains that just because AI can perform human-like behaviors doesn’t mean it can think or understand like humans.

Key Takeaways

• AI systems such as GPT-3 and PaLM can perform human-like behaviors, and Gato can perform hundreds of tasks, including captioning images, answering questions, playing Atari video games, and controlling a robot arm to stack blocks.
• Researchers at Google's AI company DeepMind claim they can produce human-level artificial intelligence by scaling up existing models.
• Most recent AI systems are built from artificial neural networks, which are inspired by the human brain. These systems can be trained to recognize patterns and generalize from results and mimic techniques humans use for such tasks.
• Neural nets are typically trained by "supervised learning", which means they're presented with many examples of an input and the desired output, and then gradually the connection weights are adjusted until the network "learns" to produce the desired output.
• The GPT-3 model was trained on 400 billion words, mostly taken from the internet. It would take a human nearly 4,000 years to read this much text.

Focus piece: “It's time to accept AI will never think like a human – and that's okay”

Executive Summary

Machine learning models make mistakes because they don't understand concepts or context, and they are easily thrown off by biased or mislabeled data that wouldn't fool a four-year-old. This Science Focus article examines the strengths and weaknesses of human intelligence versus AI.

Key Takeaways

• Technology is getting smarter, but humans are still superior to machines at computations, predictions, and recognizing patterns. Instead of viewing AI as a less-developed version of ourselves, maybe it's time to embrace our differences and learn from animals.
• You wouldn't trust a dog to give you a medical diagnosis or relationship advice, but you might trust it to sniff out explosives, assist the blind, or provide therapeutic comfort. AI can do many things that humans can't.
• Understanding the strengths and limitations of AI inspires us to leverage this technology to support people, rather than replace them, and invent new practices and solutions.

Let's unthink this!

Who We Are

Data Products partners with organizations to deliver deep expertise in data science, data strategy, data literacy, machine learning, artificial intelligence, and analytics. Our focus is on educating clients on varying aspects of data and modern technology, building up analytics skills, data competencies, and optimization of their business operations.