21923 - SCTE Broadband Aug2023 COMPLETE v1 (1)

88 Vol. 45 No. 3 - September 2023 Issue technical it had a soul and Deepmind has developed a Chatbot called ‘Sparrow’ which is claimed to be technically superior to ChatGPT [2]. The big dangers are not likely to come from sophisticated super companies like Alphabet. Smaller companies with a ‘move fast and break things’ attitude are likely to be creative and adventurous with their application ideas. But very real harms to very real people are possible with this kind of system, and these can be easily and quickly implemented by small non- expert teams (as the Koko example demonstrates). To be specific, there are at least four mechanisms of harm that LLMs such as ChatGPT can create for their users. n Firstly, they can be untruthful or inaccurate; they can ‘hallucinate’ false answers to questions convincingly and mislead users into making poor decisions n Secondly, the LLM may ‘steal’ the answer from a proprietary source, thereby potentially harming the creator of the intellectual property and the new user of the information n The third type of harm is a breach of trust. The LLM may create an impression of empathy, concern or affinity that is simply false, and by ‘lying’ in this way may cause emotional distress n Finally, using a LLM can distort and pollute information sources whether intentionally or accidentally. Therefore, given that there are applications of ChatGPT that can cause harm, and given that we can identify the causes of these harms, it seems logical that we should be able to figure out when ChatGPT can be used (or when it should not). Already, people have constructed decision trees based on their insights (see the example at Figure 4).