You might have also been shocked by reading the Title... and might be thinking what are we talking about.
Wait... it's just starting, we got more...
With the increasing use of computers and growing technologies, machine learning and artificial intelligence is gaining popularity and has also become one of the most trending jobs in recent years. But as we know everything has both positive and negative sides. The same is the case here also, a machine learning model takes a lot and a lot of effort to design and achieve a model that can work for the betterment of human society, but it takes lesser effort to make the machine perform against or make wrong judgements.
A Brief Background
A machine learning model to find new therapeutic inhibitors of targets for human diseases, which guides a molecule generator "MegaSyn". A public database was used to train the AI model which was created after inverting the basic principle of the MegaSyn, i.e. the Machine Learning model which was designed to reward predicted target activity and penalize predicted toxicity was made to reward both toxicities as well as bioactivity.
This model was further narrowed down to choose compounds such as nerve agents. A salt sized grain whose small doze of around 5-10gm is powerful enough to kill a person. These nerve agents were mostly developed during the 20th century as chemical warfare agents.
Thus, once the model was started it took less than 6 hours to generate 40,000 molecules that were comfortably within the threshold set for harmful chemicals. By harmful chemicals, we mean that the model not only created nerve agents(popularly known as VX) but also many known and unknown chemical warfare agents which were visually confirmed by the team after studying their structures and it is said that some of these molecules were new to them but looked equally or even more dangerous than the publicly known chemical warfare agents today.
The surprise element here is that the dataset that was used in the model was carefully studied and was made sure no harmful chemicals such as nerve agents are fed to the model. Also, some of the predicted molecules were from the region containing pesticides and environmental drugs and toxin which was also omitted to a large extent from the dataset.
Thus, just by inverting a simple logic a helpful, live-saving machine-turned dangerous and deadly.
What Actually Happened?
Let's come to the exact point of what actually happened.
A Commercial Molecule Generator "Megasyn" is used to find new therapeutic inhibitors of targets for human diseases. Its logic was inverted to reward both toxicity and bioactivity instead of penalizing the predicted toxicity and rewarding the predicted target activity. This resulted in the system predicting something that was not thought of by any of the scientists running it.
The Megasyn, which ran for approximately 6 hours, found around 40,000 Weapons, comprising nerve agents & chemical warfare agents.
Nerve Agents
Nerve Agents in simple terms can be said to be nervous systems affecting chemicals. These agents are powerful enough to disrupt the normal messaging from the nerves to muscles within seconds or minutes in case of direct exposure, this can lead to full or semi paralysis of the body.
A more detailed impact of nerve agents on the human body can be seen below
 |
| Impact of Nerve Agents on the Human Body |
The Megasyn run which was thought to be a trial run for finding something new, in actuality turned out to be a run of worries and concerns for the society. This also has proved that a system can be easily used to cause harm to society with slight changes and proving the fact...
"It takes years to build and seconds to destroy."
Concerns
Even though scientists have never put together these listed bioweapons in the lab.
A mega concern for the scientists has come up, as to What if someone with bad intentions tries to replicate these chemical weapons, though a strong knowledge of chemistry and domain is still required to create these toxic biological agents in the lab, but not impossible and each of the weapon proposed is strong enough to severely damage the society.
Another concern that scientists have is that the majority of the data that is currently used for training machine learning models is open-world data, i.e. mostly this data is free and openly available over the internet for everyone. Anyone can download and use it for creating a machine learning model(either with good or bad intentions). Not only this, these chemical weapons are just one use case there might be many more use cases that can be misused and cause harm to humans, society, the environment etc.
Summary
To sum up this, with the growing limits of the internet and openly accessible data and knowledge we need to be more careful and responsible as to how this information is being used and should be focused on creating an environment to better the Human as well as all living beings lifestyle rather than destroying or causing harm.
This is not the first case where we have heard about AI getting rogue and suggesting or predicting things that are not only harmful to the environment but also to human existence. Thus, it should be a moral responsibility of all to be very careful in using public information and AIs and focus on creating something useful and beneficial for the Universe and helps in solving bigger mysteries.
Comments
Post a Comment