SynTag

Why is AI customer support far from solved? We know AI excels where humans can fall short. They don’t sleep, stress out or forget anything at all. We also know that customers love personalized attention, and hate waiting in call queues. These are both issues solved with AI bots. So with the rise of AI CX solutions like Intercom and Bland AI, why is AI customer support still far from being solved? We can think of at least four reasons: 1. Exploitability: Customer support relies on (large) language models, which have been shown to be “hackable” (see https://lnkd.in/e9PEg8E4, a deep-dive into LLM vulnerability by Learn Prompting). This can be mitigated with a combination of severe topic restrictions and disclaimers, all of which contribute to longer pauses and slower conversations. 2. Still needs to be programmed: A popular antidote to hallucinated answers are programmable control flows (Twilio, Bland AI). Those use if and else statements evaluated by an LLM. While there are many real-world use-cases that can be modeled using if and else statements, it still lacks the fluidity of a natural conversation. 3. Inefficiency: While bots save businesses valuable time answering basic questions, conversations must be checked for accuracy after the fact by humans. 4. Lack of human touch, or “understanding”: True understanding and empathy has yet to be achieved between the bot and the human. These valuable factors are needed to elevate these bots to a level of trust with the consumer. Conversational AI will say the most likely next word. This is determined through a probabilistic calculation. They don’t think in mental concepts. They have no concept of empathy. SynTag is building voice AI to overcome these challenges. Are there any other challenges you can think of?