The artificial intelligence revolution has brought many open-source AI tools to the public attention. Many of us have used tools such as ChatGPT, MidJourney, or Canva’s suite of AI services at work or for fun, while other AI applications hum away in the background of our smartphones, home assistants, and automobiles. And yet while many of these tools are becoming increasingly familiar to us, questions remain. How secure are they? What are the ethics of using generative AI in professional settings? And how suitable are they to highly-regulated industries like life science?
In this article, we’ll compare some of the most popular open-source AI solutions, assess the suitability of these tools for life science applications, and explore the potential of AI tools designed specifically for the life science sector.
OpenAI
OpenAI is the architecture that powers ChatGPT, the image generation tool DALL-E, and others. ChatGPT excels at generating coherent text from human prompts, giving it numerous professional and casual applications. These tools can be customized for different industries or requirements using a variety of APIs and plugins, and integrate with a number of third-party software tools – such as Microsoft Office, and project management tools like Trello and Asana.
OpenAI tools provide excellent support for writers, programmers, and educators, and can also underpin chatbots to deliver high-quality customer service.
Google Gemini
Google describes Gemini – formerly known as Bard – as a “personal AI assistant”. It’s effectively a superpowered chatbot that integrates with the Google search function to provide highly-contextual, up-to-date responses. As the tool is supported by Google’s technologies, it performs well across languages – allowing for multilingual communication and document translation.
Gemini is adept at summarizing text and providing prompts to where copy can be refined, while Google integration allows users to research with live data. Like OpenAI, Gemini can be a powerful tool to support writing and education projects, and can also be used as a chatbot in customer service applications. But unlike OpenAI, Google Gemini combines its text, image, music and video tools within a single interface.
Claude
Claude is a large language model developed by Anthropic, with the promise to “help you do your best work”. Safety and accuracy are watchwords of the Claude AI assistant – it’s been designed to address some of the concerns around reliability and data security common with open-source AI.
Claude features a large context window that allows for longer text prompts than many of its competitors, and features an intuitive interface for ease of use. As such, it’s great for analyzing large, in-depth documents such as legal and regulatory files, and as a conversational AI in sensitive domains where privacy, accuracy, and security are especially critical.
Microsoft Copilot
Microsoft’s Copilot AI is designed to integrate seamlessly with Microsoft Office tools and Microsoft Teams – adding powerful AI functionality to optimize your workflow. It offers features specifically designed for business applications, such as data analysis and report generation – making it especially suited for corporate environments.
Copilot’s strengths include corporate report writing, meeting transcription and summarization, and spreadsheet analysis.
An AI comparison chart
This comparison table is a useful reference guide for the relative strengths and weaknesses of these common open-source AIs:
Open-source AI and life science
While open-source AIs are incredibly powerful tools with numerous useful applications, on their own they aren’t suitable for use in life science. There are a number of reasons for this:
Data privacy/security concerns
Pharmaceutical, medical device, and other life science companies handle a lot of sensitive information. Patient data, claims data, clinical trial data and more must be kept secure and private – it can’t be shared in a public domain like an open-source AI tool. As a result, many life science companies are understandably reluctant to integrate AI tools into their tech stacks.
Compliance issues
Life science is a highly-regulated industry. Most open-source AI tools fail to comply with these regulatory standards, while many life science companies have their own policies regarding artificial intelligence that further restrict the use of open-source AI tools.
Lack of data specificity
Lastly, open-source AI tools are trained using very large, broad data sets. While this is ideal for more general or business applications, they lack the specificity and deep industry knowledge to be applicable to pharma strategies.
“What we’ve found is that a lot of off-the-shelf large language models tend to be very broad, and they’re trained on a lot of broad data. They’re trained to be very smart in their understanding of the text, but they’re not specific to what we do in life science and medical affairs.”
– Jason Smith, Chief Technology Officer, AI and Analytics, Within3
The Within3 approach
Open-source AI tools are powerful, and growing more so all the time – but they’re effectively off-limits for life science insights gathering. Within3 has addressed this issue by building on existing large language models like those supplied by OpenAI and Anthropic – layering on our own proprietary technologies to specifically address the needs of life science companies.
- Addressing data privacy and security concerns: We don’t use customer data to train our AI models, we don’t retain our clients’ sensitive information, and we don’t share proprietary data with open-source tools like OpenAI.
- Addressing compliance issues: We have our clients pre-authorize the use of any potentially-sensitive data, and don’t touch highly-sensitive information like clinical trial or patient data – thereby helping our users remain compliant.
- Adding data specificity: Our AI is highly specific to life science, and uses customer data to generate insights that directly inform your strategy. Our AI pulls data from your CRM and enterprise data lake, congress data, and various external sources to provide contextual insights and recommendations.
“We still leverage pieces of off-the-shelf technology, but what we do is sandwich that technology. We have our unsupervised learning models and technologies that evaluate text to find patterns in the data, then we can be more specific about the use of off-the-shelf large language models to summarize that information.”
– Jason Smith, Chief Technology Officer, AI and Analytics, Within3
Within3’s AI is built for life science. We’ve developed far beyond generic, open-source AI tools to create a system that truly understands your strategy – one that speaks pharma, and is capable of generating digestible, actionable insights that inform strategic decision-making.
Want to find out more about our AI? You can read about our approach to artificial intelligence in life science here, or to experience it for yourself, you can schedule a demo today.
