With artificial intelligence the possibilities are endless. AI and its skill set are continually expanding. Drugs are only the latest goal.
These machines, models and algorithms become more human every day: they basically compress the decision-making and discerning power of an active human brain into codes.
The results are already beyond our limited capabilities. Artificial intelligence as the future of work and productivity is an understatement: it could be considered as THE future, in a general and extensive sense.
AI can now “prescribe drugs” – experts believe this will have a positive impact on overall health.
Research into AI's ability to "prescribe drugs"
The research shows that drug-related errors are responsible, only in the US, one in 131 outpatient patients and one in 854 hospital deaths. The act of prescribing drugs is as complicated as it is vital, which medical experts must practice years before mastering.
Artificial intelligence, however, is on the verge of replicating these capabilities with machine learning, and making prescribing as easy as clicking a button.
Another example of artificial intelligence beating humans at their own game? For example, artificial intelligence models can already “designing” oncological therapies, or provide for any complications in pregnancy.
But how can they do it? How can AI models prescribe drugs?
GPT-3: So good it's scary, literally
The internet has gone crazy for GPT-3 (Generative Pre-Training Transformer-3), an artificial intelligence developed by OpenAI. It's a third generation machine learning model that designs websites, answers questions and, yes, can prescribe drugs.
Its “big brother” GPT-2 was already considered dangerous
OpenAI has set the AI world ablaze with security debates in the past; the introduction of GPT-2 was fulfilled similarly.
Although GPT-2 already brought benefits, critics called it too much dangerous. He could create writings indistinguishable from those written by real humans. He was TOO good.
Consider that GPT-2 used only 124 million parameters of the possible 1,5 billion in its original design. Well: GPT-3 will feature a staggering 175 billion parameters. If GPT-2 was “dangerous,” what should we think of its successor?
GPT-3, wide open to the phenomenon (even with drugs)
GPT-3 is a neural network-based language model that predicts the probability that a sentence exists in the world.
GAN
GPT-3 exploits an adversary generative model said GAN (two neutral networks perfect each other for competition). Can acquire knowledge and process long-range dependencies on different sets of written material with long strokes of contiguous text.
As a language processing model with the largest database of training sets ever at its disposal, the experts they believe it can answer questions related to medicine, diagnose asthma and prescribe medications.
Prescription drugs: Google also has its own AI
As powerful as GPT-3 is, it is not the first AI-based model capable of prescribing drugs. Google's AI could provide the recipes that a doctor will write with an accuracy of up to 75%.
There are several unfavorable points in this model. The most significant setback of Google's AI is that it is rooted in historical data. It can only replicate your doctor's prescribing patterns and not their ever-expanding knowledge of medications and side effects.
Based on the researchers' presentation, however, the system applied appropriately to healthcare could assist doctors in identifying abnormal or incorrect prescriptions.
It would work similar to the fraud detection programs used by banks.
In other words, when prescribing drugs, artificial intelligence would be a useful tool capable of reducing the error rate of the human doctor, who is still unbeatable in this regard.
For how long?
Bianca Stan – Graduated in Law, writer with several books published in Romania and journalist for the group "Anticipatia" (Bucharest). It focuses on the impact of exponential technologies, military robotics and their intersection with global trends, urbanization and long-term geopolitics. He lives in Naples.
