3 ms·
The State of Voice Assistants: Where's the AI?
- neuronDen 7y agoReposting this post from 2018. Most of the issues still remain unsolved in 2019. Looking forward to the comments from the HN community.
- streetcat1 7y agoThe issues are there from the 80's (NLU). The only advancements made in this field are speech recognition (which reached human level, and thus help in reducing error rates at the input to the NLU component), and maybe deep RF, to find the most optimal conversation response, however the machine still do not have a semantic understanding of users utterance, and will likely never have. I.e. if you solve the NLU issues, you are likely very close to AGI.
- neuronDen 7y agoYou don't need NLU to be fully solved for assistants today which work on narrow range on tasks. For example - modern deep learning techniques can easily mark "Send a message to <person> - What are you upto?" and "Check with <person> what he is upto" as similar. Assistants today respond correctly to the first input but not the second.
- streetcat1 7y agoSo I am talking about spoken dialogue system. I.e. system that can converse (I.e. at least 15-20 turns between the user and the system).For example, a system that can book multiple legs flight.
- neuronDen 7y agoMicrosoft acquired Semantic Machines has the capabilities to do so. http://www.semanticmachines.com/ http://www.semanticmachines.com/
- streetcat1 7y agoAs I said. This is a problem from the 80s. Why would it would be solved now? and by a specific startup? what changed?. I think that in order to solve this issue we would need some breakthrough in symbolic AI and not just statistical AI. However, most of the research today is done on DNN and their derivative.