What Is Generative AI?
Generative AI
Generative AI tools can create new content, such as text, computer code, images, audio, sound, and video, in response to a user’s prompt, often in the form of a short written description of the desired output. Generative AI tools are based on machine learning, trained using enormous amounts of data.[1] Generative AI tools are built on a system of inputs and outputs. First, the tool goes through a machine learning period whereby it is trained to generate predictive models and creative outputs through a large data set, often varied and diverse but tailored to the goal of the tool (i.e., customer service, generating scientific or marketing models, etc.). For in-house tools, this can be done with the company’s own data; for larger tools such as ChatGPT, this is done with the creator’s data set.[2] Once the tool has been trained, the individual user “inputs” a short prompt for the tool to synthesize and produce an “output.” Inputs are often retained on the servers controlled by the company that supports the tool, for monitoring of the tool’s performance and, in some cases, continued learning. The “outputs” are created by combining the machine learning during the training period with the inputs to produce an output.[3]
Even though Generative AI has only recently become widely available to the public, these tools are already routinely being used in the workplace. Recent studies have found that 56% of workers have tried using Generative AI in the workplace, with nearly 1 in 10 employing the technology on a daily basis.[4] Another study found that sensitive data, which may include the company’s own competitively sensitive data or client sensitive data, comprises up to 11% of what employees paste into the tool.[5] Even more troublingly, source code was the second‑most common type of confidential data provided to ChatGPT in the six-week period studied.[6] However, AI tools can be programmed in various ways to avoid disclosing certain types of information.[7] This is both promising, in that protections are possible, and troubling, in that there are many ways that trade secrets may be implicated using Generative AI.
Winston & Strawn has developed a dedicated multi-disciplinary AI team with extensive familiarity with the trends and regulations in the rapidly changing area of AI.
[1] IBM, What is Generative AI? IBM Research Blog (Apr. 20, 2023), https://research.ibm.com/blog/what-is-generative-AI.
[2] Stephen Amell, How to Train a Generative AI Model, Medium (June 16, 2023), https://medium.com/@iamamellstephen/how-to-train-a-generative-ai-model-1ab605615acd; supra note 6.
[3] Id.
[4] The Conference Board, Press Release: Majority of US Workers Are Already Using Generative AI Tools – But Company Policies Trail Behind (Sept. 13, 2023), https://www.conference-board.org/press/us-workers-and-generative-ai.
[5] Cyberhaven, 11% of Data Employees Paste into ChatGPT is Confidential, https://www.cyberhaven.com/blog/4-2-of-workers-have-pasted-company-data-into-chatgpt/ (last accessed Nov. 28, 2023).
[6] Id.
[7] Jeremy Elman, AI and Trade Secrets: A Winning Combination, IP Watchdog (Nov. 28, 2023), https://ipwatchdog.com/2023/11/28/ai-trade-secrets-winning-combination/id=170001/. (“Even today, some of the commercially available Gen AIs are implementing a proprietary mode where companies can designate some of the information output as confidential already, so “reasonable means” for keeping inputs and outputs confidential appears possible and qualifying as a trade secret. This would simply be an implementation detail in the way that the information is disclosed. Companies can then license these trade secrets to prevent the output from becoming public.”).