At Hint Summit each year we strive to push the boundaries in terms of both new ideas within our space, and ideas and stories from tangential fields that can expand and open our minds. So this year, we were excited to invite Sean Gourley, Founder and CEO of Primer AI, to deliver a keynote on ‘The World of Artificial Intelligence’ at Hint Summit 2022. “Artificial intelligence is going to have its biggest impact in the places where human intelligence has its limitations. And these human limitations are hardwired into our system and we can’t change them.”
The human brain cannot respond faster than .1 seconds while artificial intelligence can. There have been many developments with artificial intelligence, however it is still very much in the early stages. A progression of deep learning has taken place with AI and some highlights include speech recognition in 2011, image recognition in 2015 and natural language processing in 2019.
The development of deep learning came about by providing more hidden representations, which is in between the input data and output. There are three major events that led to a more efficient AI. This includes ImageNet, which was a project that took millions of hours of human labor in identifying and labeling 14 million images deployed for machine learning; the growth of GPU computing power; algorithms or neural networks. Because of these developments, AI went from a 30-40% error rate to a 5% error rate in just 3 years.
Sean points out important breakthroughs in the evolution of AI including in 2014 when machines beat humans at street sign recognition 99.2% vs 98.8%. Now machines can not only just recognize images, but also describe them accurately. More recently in 2022, machines beat humans again with image captioning at a score of 138.2 vs 125.5. However, when it comes to imagination, humans still have the upper hand because unless a machine has been trained to recognize something, it gets it wrong.
In 2015, Generative Adversarial Networks challenged the machines to create images based on text descriptions and they were able to generate fairly accurate images. Additional computing power allowed machines to generate more realistic and hi-resolution images over the years. Sean then demonstrates the creative ability of machines with examples of machine-generated artwork.
The next layer of deep learning for machines is language, which is an area that Primer AI is most focused on. Sean shares examples of AI writing headlines and scripts. There is an immense opportunity for AI to get smarter but through that learning curve, there will be catastrophic failures since AI makes decisions within set parameters.
In the medical field, there are instances where providers can leverage AI to enhance the accuracy of diagnoses. Sean cites a study by Wang et al (2017) comparing how the AI and humans detected a tumor. The AI had 92.5% accuracy while humans had 96.6% accuracy but when combining the two, a 99.5% accuracy in tumor detection was achieved. Sean believes that it isn’t about AI and humans competing when it comes to intelligence, but rather bringing together “the best of machine intelligence with the best of human intelligence to solve the most pressing problems of our time.”
Watch the full keynote to get additional insight on AI as well as a thought-provoking Q&A session, led by Hint Health's Co-Founder & CEO, Zak Holdsworth.