ChatGPT: weighing the risks & benefits

ChatGPT. A few days ago, its ex co-founder Elon Musk termed it the “most dangerous” #AI to date. Moreover, JPMorganChase , Amazon , Verizon and Accenture staff reportedly have been barred from using #ChatGPT for work. BNP Paribas is no exception.

As individuals, as students, as teachers, as coders, as managers, as marketers, … we are convinced that GhatGPT is of great help. Indeed, it’s a super useful and incredibly powerful tool that also allows an amazing gain in productivity  ➡️ So why companies are not eager of using it internally? 

 Well, while ChatGPT offers many benefits, its use also poses several security risks ⚠️ that organizations must be aware of to protect their sensitive data and reputation. 

1️⃣  One of the primary concerns with ChatGPT is the risk of data breaches. In fact, the quality of the answer proposed by ChatGPTdepends on the accuracy of the request and the information provided which may include sensitive information such as personal information, financial #data, and confidential business information. Indeed, the quality of the prompts given to ChatGPT directly affects the accuracy and relevance of its responses, ultimately determining the overall effectiveness and usefulness of the model  ➡️ So by using it companies could run a risk of data breaches. If this data falls into the wrong hands, it can cause significant harm to the organization, including economic loss, reputational damage, and legal liabilities 📉

2️⃣ Another risk is the cyber one🧑💻. The main concern here is the fact that insofar as programs can be established from code provided by ChatGPT, these "bits" of code would not be confidential and could be used as an entry point by malicious third parties to "enter" programs within companies. Another risk would be to use code snippets provided by ChatGPT which could include viruses, pirate sensors etc.

3️⃣ Let’s not forget that ChatGPT is a new form of interaction based on data it can access on the web. At the end of the day it is a pure statistical tool even if amazingly powerful. This statistical base is known to be biased, so might be the answers, and its algorithm doesn’t provide sources guiding the users unlike search engines. Therefore, even if #LLM (Large Language Models) add control rules to avoid going into subjects that would be too risky or too sensitive, we as humans must learn how to use it, how to cross-reference information, to look for sources, to reorganize decision processes… and it’s a too early stage for a responsible company like ours. Like this very interesting article here outlines, LLM are sufficiently revolutionary to alter our understanding of language itself. So, we must learn how to control them, how to supervise them and understand what role they can play in language development, as simple octopus or "stochastic parrots » (see article on Emily M. Bender work).

4️⃣ Another important issue with ChatGPT - energy consumption⚡️. As some sources indicate, it may have consumed as much electricity as 175,000 people in January 2023. It’s too early and too complicated to make a right estimate at this point of time but should be considered while being deployed.

5️⃣ Finally, ChatGPT's conversational capabilities can also pose security risks such as style imitation or scam. Cybercriminals can use the ability to simulate human-like conversations to deceive users and gain access to sensitive data or networks.

☞ Therefore the decision to bar ChatGPT for work in our organisation is the result of a thorough consideration led with our cybersecurity teams. For us as a financial group🏦, data is key so it’s our responsibility to navigate innovations and mutations in their whole to ensure that data safety is enforced.

✓ That being said, I am fairly certain that LLM as their whole will play a key 🔑role in the future of our activities. They have the ability to raise productivity in a wide variety of services and also to transform information management, knowledge management or customer information processes as we know them. ChatGPT is a powerful tool that can be used by businesses to provide customer service quickly and efficiently. It can be used to answer questions, provide customer support, and even generate content for websites and blogs. It is also used in chatbots, virtual assistants, and other AI applications.

As we say, “better be safe than sorry”, and as a company where #cybersecurity is a cornerstone, innovation must be dealt with at a pace that doesn’t put us at risk. I am confident that as an example on premise LLM will expand in the next future.

Be ensured that, as a company where #technology is one of the three pillars of our strategic #GTS plan, there’s no doubt that we will keep looking into it to take the most of it.