Artificial Intelligence is not intelligent, and Machine Learning is not Learning…
Artificial Intelligence is not intelligent, and Machine Learning is not Learning…Well, at least not yet… It does seem that the time when robots are taking over the world is right around the corner. Cars are driving themselves. We talk to objects, and objects are responding back with smart answers on any topic (with exception of health-related topics). Machines can write articles so well you would never know it was not written by a human: https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3 It’s all swell and exciting. However, at least for now, the AI is as good as the human brain who has created it. I want to take a deep breath, forget about the hype and theories for just 1 minute and analyze what we have right now and in the near future. Most importantly, I want give credit to the amazingly smart humans who are building this great technology. It’s humans who are running the show right now, not the robots. Just look at the record high number of jobs for data scientists and machine learning experts. Artificial intelligence is not designed to take on tasks it was not meant to. Roomba can vacuum. It could vacuum very well. It could potentially use deep learning to vacuum even better, faster and with higher quality. However, unless programmed to do so by humans, it’s not going to “learn” to do other tasks. Machines are great at routine, repetitive tasks. They are great at math: computing and aggregation. And they are super-fast. But for now there is no one better than humans in humility, social skills and intuition. These are important characteristics anywhere, but especially in healthcare, when you are often dealing with life-and-death situations. Another thing is data. Most of the world data is garbage, a virtual garbage: https://www.technologyreview.com/2021/04/01/1021619/ai-data-errors-warp-machine-learning-progress/ As of now, in 2021, the best machine learning technology is making errors even with “good” data, as in the panda-vs-gibbon example: https://towardsdatascience.com/breaking-neural-networks-with-adversarial-attacks-f4290a9a45aa With the “dirty” data, which is majority of the world data, how would one expect AI to learn? Data scientists are aware of these issues. I believe humility is what distinguishes excellent data scientists from the world class data scientists. What we’ve noticed in our own WellAI research is that even textbook model validation on “clean” data could be very different from real life. The latest and the shiniest algorithms could suffer from ‘overparameterization’ and ‘overfitting’. Simpler is often better, as the MIT Technology Review article correctly points out. As for the data itself, quality of the medical data is especially important as we are dealing with people’s health and people’s lives. Ironically, healthcare suffers from erroneous and biased data, partly because the so-called Big Pharma and Big Tech are profit driven and could bias medical and clinical studies in certain direction, often in a very subtle, unnoticeable way. In WellAI’s educational live sessions, we very often emphasize the importance of humility and not taking any AI-generated outcomes for granted. For example, in this short clip of the lecture given in February 2021 my co-author, WellAI’s partner and colleague in the IFCC’s working group on AI & Genomics Daniel Satchkov talks about being humble about AI: https://www.linkedin.com/posts/spolevikov_pandemic-health-ai-activity-6767101066002743296–mHf Daniel points out that both the terms AI and ‘learning’ are complete misnomers and quite apart from data problems. These algorithms are not intelligent and do not learn. Adversarial examples such the panda-vs-gibbon one mentioned above have put to rest any generalized learning pretense long ago. Data problems described in the MIT Technology Review article must be solved. That is a given. But the bigger problem is overreach of pitching these algorithms as being able to learn and solve whole classes of problems that they simply cannot. Machine learning is great for extracting patterns from data and human behavior, and it could be a great help in automating some important tasks and freeing up humans. But beyond that, it is hype and snake oil. So sleep well. Skynet is not sending a Terminator to kill the human race. At least not yet… Don’t get me wrong – I am super excited about AI. Heck, I am running an artificial intelligence company and loving it. I do believe there is a place for AI, and there is a place for humans. And the best and most productive scenario (for the healthcare system in particular) is when the AI and the humans are working together, utilizing their own specific skills and talents. At WellAI, we use artificial intelligence for what we believe artificial intelligence is supposed to be used: quick, efficient, cheap and accurate information access to the knowledge of 30+ million medical studies. Try WellAI for yourself. The price for our flagship voice guided digital health assistant is still super low at $39.99 per year (which is equivalent to $3.33 per month). You can make your purchase directly on our website https://wellai.health or at the App Store or on Google Play on your smartphone. If you want to learn more about this break-through technology that is meant to provide unique access to scientific health information to everyone and to dramatically improve efficiency of your health spending please schedule a 1-on-1 demo:
- • If you are an employer, a doctor, a health benefit advisor, or an insurance broker and would like a cool benefit that makes your employees healthier and happier, and saves 34% to 45% on your healthcare costs, please schedule a virtual meeting here.
- • If you are a potential investor who would like to be part of a unique once-in-a-lifetime investment opportunity that will forever revolutionize healthcare, please schedule a virtual meeting here.
- • If you are a potential customer, a patient or you are just curious about any of the WellAI high tech products – ‘Health Alexa’, next gen telehealth, or the scientific chronic disease management program – please schedule a virtual meeting here.