Impact of AI in biology
Posted on October 27, 2021 • 5 minutes • 902 words
Welcome back, dear readers. I’m back with a friend of mine. After a good discussion, we come to you with a brand new interesting topic, “Impact of AI on biology”. AI, Machine learning, deep learning, these words have become quite prevalent in the latter half of this decade. Several famous authors with a huge outreach such as Yuval Noah Harari have expressed their ideas on AI, mainly on its potential to completely revolutionize the path down which our civilization heads. On a lighter note, we’d like to share our knowledge on the impact of AI on biology and our views of the future of biology impacted by AI.
Inspired by the Neurological complexity and biological evolution, deep neural networks or recurrent neural networks drive the learning process and its refinement in algorithms. Artificial Intelligence tries to replicate behaviour associated with humans, such as learning, pattern recognition and problem-solving. AI and Machine Learning have been able to match and even greatly surpass human cognitive abilities in recent years. Now, cognitive computing has evolved to make its own decision. One of the most complicated cognition of humans is to process vision and language. The winner of the recent ImageNet challenge has shown visual cognition with a top error rate of 3.8% while human accuracy is around 95%. Machines are now able to learn and utilize languages such as English, Sanskrit or character-based languages such as Mandarin, Japanese Kanji, Egyptian hieroglyphs and other variables as demonstrated by Hahn and Baroni (2019).
The impact of computational biology on the field of medicine is fast growing. Machines can now diagnose diseases, track their progression and suggest multiple treatment options. In a more day to day implementation of AI and Machine learning, most of us have come across fitness tracking apps and accessories. These devices collect data from one’s BMI, daily activities, food habits and other unstructured data and suggest potent solutions in the prevention, management and treatment of ailments such as obesity. Furthermore, wearable accessories under research have been trained in detecting heart health and potential risks accurately. Hatib and their group trained machines to detect the incipient onset of hypotension up to 15 min before hypotension actually occurs.
Machines can now analyze your pathological reports and suggest the most appropriate course of treatment. It is rather amazing that machines can discover newer subtypes of diseases unknown before, as seen in asthma patients, advising a different course of treatment. The use of AI in medical imaging (such as EEG, ECG, biopsies), has shown tremendous potential to accurately and quickly diagnose and identify the prognostic markers in a patient and suggest the course of treatment more wisely. With appropriate data, it is also possible to identify the disease before its occurrence.
Just by using Smartphone pictures, cognitive algorithms can currently detect the type of wounds or external injuries or snake bites, reaching a specificity of 98%. In surgical procedures, ML is applied mainly in 2 areas, namely: Robotics and decision support. Robots have now performed and assisted surgeries including hair transplantation, laparoscopy and minor cardiovascular surgeries. Machines are not only highly capable of suggesting a perfect therapeutic strategy; they are now intensively used to develop novel therapeutics. Halicin is a good example. It’s a novel broad-spectrum antibiotic identified using deep learning. The company distributed bio is working towards developing a single vaccine against flu, or a single antitoxin for all kinds of snake venom, just by running computers. Many companies like Arzeda, Alphabet and Microsoft provide machine learning programs that are equipped to identify and build structures of novel drug targets, analyze the safety of drugs and design experiments to evaluate the safety.
Machine learning and AI is extensively used in geographical and environmental safeguard and study. Neural networks have been utilised to build the tree of life, relation between all the species of the world aiding in biodiversity and conversation aspects. California, being highly susceptible to forest fires, has utilised both big data and machine learning to take preventive measures against potential fires and has drawn safety measures for public safety and minimization of damage.
Even though machine learning shows such tremendous potential, on the application front it suffers from quite a lot of setbacks. One such example is the faulty interpretations carried out by the QRisk2 assessment tool that has given incorrectly constructed cardiovascular risk scoring, requiring many thousands of patients to be recalled and reassessed. It is important to note that despite the presence of seemingly large enough medical data sets and adequate learning algorithms available for many decades, thousands of papers applying machine learning algorithms to medical data, very few have contributed meaningfully to clinical care.
But again, we end this blog with a possibility of a future where the requirement of doctors can be completely ruled out and algorithms aided by researchers and developers can exponentially outperform and drastically reduce costs in comparison to conventional methods. Adding a clause to our previous statement, when IBM Deep Blue has defeated Garry Kasparov in chess, all the grandmasters of chess did not simply cease to exist. Similarly, an algorithm will not turn all the doctors obsolete now, but rather will eventually fill in and eliminate the need over a period. Time being a mystery; this can happen in decades, years or possibly in months.
I thank Aman for co-authoring this blog post with me. He has contributed in synthesizing the words and flow of ideas.