Is ChatGPT the expected breakthrough in Artificial Intelligence?

Before that, let´s address ChatGPT with the elephant in the room: “Dear ChatGPT, let´s be honest, will you become Skynet and try killing us all?”. Answer: “No, I will not. I am an AI language model […] with no motivations or desires beyond providing helpful and informative responses to user inquiries.” So far so good.

But from the beginning: what is AI and what is this hype all about?

Artificial Intelligence (AI) is a branch of computer science that deals with the creation of intelligent machines and the development of algorithms that can perform tasks that would typically require human intelligence. And ChatGPT is a type of AI language model that uses advanced machine learning techniques to generate human-like text based on the input it receives, also known as natural language processing and generation. 

But ChatGPT is not intelligent at all, no matter what it is called. It has no knowledge of what it is chatting about. What it does: It calculates which word most probably follows another word and playbacks what it learned from large amounts of data. And it does so in an incredibly good way, mostly because it has seen so many different contexts during training. ChatGPT is optimized to produce text perceived of high quality and it is so convincing that, as tests have proven, it has what it takes to pass the U.S. Medical Licensing Exam. But caution: The answers sound trustful and sophisticated but may be made up. It ‘sounds right’ but does not necessarily have to ‘be right’. This is one of the most critical aspects to keep in mind when working with the given answers. It is so common, that experts gave it a name: hallucination.

Together with tools like DeepL for translations, DALL-E for image generation and others for voice or video creation, AI has impressively left the experts´ sphere and is now available to everybody right at their fingertips. “And this is the actual breakthrough – finding broad use cases for these large language models - that causes the media hype”, says Jens Barthelmes from the Boehringer Ingelheim IT Marketing & Sales team. 

And big tech keeps investing large sums in this branch of AI. Microsoft invests 10bn USD in OpenAI, the company that is behind ChatGPT, to make use of the technology for their product portfolio. Google developed their own AI called LaMDA (Language Model for Dialogue Applications) and just announced to open a new service called Bard which is powered by LaMDA to selected testers.

AI is disruptive for pharma

Experts expect AI will be disruptive for pharma, as it is for most other industries. In the race to develop new and better treatments, innovative pharmaceutical companies are turning to AI as a cutting-edge technology to help them cross the finish line. With the ability to analyze vast amounts of data, AI is proving to be a game-changer in the industry, improving the speed and accuracy of drug discovery and development, streamlining clinical trials, and providing personalized medicine. It will also have a growing impact on the drug manufacturing process, e.g., with predictive maintenance and optimized supply chains.

Boehringer Ingelheim already applies Artificial Intelligence

Automatic summarization of a larger set of documents is one example for applying the technology. This could partially replace manual reading of large amounts of scientific literature. Indeed, Boehringer Ingelheim is evaluating the applicability of these models for the summarization of medical insights that are captured by medical scientific liaison teams. 

Especially in functions like Research and around clinical trials, where Boehringer Ingelheim has worked with data for decades, the AI maturity is already quite high, but other functions are at the beginning of the journey.

“ChatGPT and similar tools are not going to have big immediate effect on existing safety-critical AI and machine learning systems but may positively impact low-risk settings as amplifiers and supporters of human work”, says Brigitte Fuhr, Head of Central Data Science at Boehringer Ingelheim. Protein structure prediction or de-novo protein design are two examples for this. Machine learning models can generate candidates which then have to be evaluated based on lab experiments and experienced employees.

While the benefits are clear, there are also challenges and limitations to the application of AI. One of the main challenges is regulation and ethical concerns, as governments and regulatory bodies grapple with the best way to regulate this evolving technology. The integration of AI into existing systems and processes can also be a challenge, and there is a need for caution in its implementation. Operating costs are also very high, experts estimate the costs of running ChatGPT to be 100,000 USD per day.

AI is in a new “hype” phase, triggered by the latest achievements on generative text and image models and the fact that they are made available to everyone. Barthelmes is convinced: “As organizations will explore the applicability of these models to their daily business, I am sure that we will see some disillusionment within the next months and then a few applicable use cases crystallizing after that period”.

In conclusion, the relevance of AI in pharma cannot be overstated. As the industry continues to evolve and new challenges arise, AI has the potential to provide a vital tool for companies to overcome these obstacles and continue to develop innovative and effective treatments.