The latest version of Open AI’s leading large language model tool has already made a dramatic splash in the world’s media since its launch in November last year. Indeed, ChatGPT is promising to transform the way that many organisations across a variety of industries and sectors can carry out their operations.
The Benefits of ChatGPT And The Potential Pitfalls
The benefits that ChatGPT4 can deliver are certainly impressive. The AI powered chatbot tool can not only enhance the customer service experience, but it can also provide useful insights that will help businesses to make better decisions. It can create content, easing the burden on marketing departments, and also produce reports and written data analysis. It promises increased efficiency with an improved standard of accuracy.
The data collection and analysis capabilities of ChatGPT are perhaps the most valuable for businesses, but it is here that potential drawbacks start to emerge. One key concern is the nature of machine learning, which requires access to, and analysis of, so-called “training data”. From a privacy perspective, there is the question of where this data may be drawn from. This has already been illustrated by Microsoft, one of Open AI’s major investment partners, telling its staff to be cautious about sharing “sensitive data” with the chatbot lest it end up being subsequently used in this way [1]. Amazon, has similarly expressed concern about this issue [2].
Don’t Forget Compliance
Of course, even as the technology continues to evolve, the organisations seeking to use it are still bound by strict compliance regulations, which relate to data privacy as well as the way in which it is stored, managed, and accessed. With large language models such as ChatGPT, there is a very real chance that PII is being collected for use as future training data. This could be problematic for individuals later trying to exercise their “right to be forgotten”, as well as raising more general security concerns regarding storage and access to such information.
Furthermore, under current GDPR regulations, should an organisation use ChatGPT’s functionalities in such a way that they come under the GDPR category “Profiling and Fully Automated Decision Making”, then the data controller could find themselves in complex legal territory. GPDR compliance challenges could also arise from the fact that Open AI reserves the right to take data collected through ChatGPT chatbot use and store it in the US for training purposes.
Is There A Solution?
Breaching GPDR regulations comes with severe repercussions, both in terms of fines and potential legal action, but also in terms of reputational damage. So, how should businesses wanting to take advantage of the power of AI move forward?
In the first instance, following the example of Microsoft and Amazon would seem a wise move. Train staff so that they know not to share personal or sensitive data with the technology. If including personal data is essential, then ensure that the right consents are acquired. Have a human pair of eyes check over any documents generated by the tool, and eliminate any potentially problematic data. Bear in mind that the nature of machine learning can mean that generated outputs can be affected by biases present in the original training data, which, in the case of racial, gender, or discriminatory content, can also be highly problematic.
AI Tools – The Discussion
The question of meeting GDPR and privacy requirements whilst benefiting from AI is clearly a hot topic and explored in detail in the Morpheus Meets Episode 4 podcast where Daniel Ben Tov (VP of Engineering at Hyro) talks to Morpheus’ Sean about his views on the issues. He explains that large language models are getting better and better, and, whilst overcoming data privacy and compliance issues will be an “interesting journey”, the tools will ultimately deliver a service which is “10 times better” for companies. We’d love to hear how your own experiences of using this technology compare – do you agree with Daniel’s findings?
To watch the full episode – click here
Resources:
