重新思考训练AI的方式

当艾伦·图灵提出“机器智能能做些什么?“的假设时,他的疑问实际上在逐渐发展为一个更加实际且可操作的问题形式-从“机器可以思考吗?“发展到“机器可以完成人类能完成的事情吗?”

When Alan Turing postulated what machine intelligence could do, his question gradually evolved into a more practical and implementable form – from ‘can machines think?’ to ‘can machines do what we can do?’


Fast forward to present day, the combination of databases and technology has made machine learning more relevant in solving multiple business challenges. Rapidly digitised financial data offered a glimpse of what machine learning and AI can deliver, and the financial industry has become a hot bed of fervored talk on using AI to create new profit centres and reduce operational costs.

 

Yet businesses still grapple with the promises machine learning and AI, often wondering if expectations are inflated or if digging deeper and persisting could reap more impactful returns. I’d like to think that it is the latter.

 

With continual refinements in machine learning techniques, AI implementation should be more orchestrated in its machine’s content training strategies. Having a well-defined business metric is only the start. But to stop there and generalise machine learning to “letting the algorithm handle all the data” may not be the most effective strategy to achieve significant outcomes.

 

A cursory understanding of the building blocks of machine learning – supervised, unsupervised and reinforcement learning – may suffice from a business standpoint. But deeper appreciation for the statistical concepts that underpin these blocks hold the key to developing content for more robust machine training strategies.

 

Each business case has a unique mix of statistical concepts. Familiarity with knowing what backs up the machine’s thought process gives the business more options and to craft training strategies that support the machine’s learning, particular in the case of chatbots and natural language processing (NLP).

 

For most of us, the simplest way to train a chatbot is by pairing an answer to each question. You might go further and pay more attention to frequently asked questions. However, machines just like humans, need context to understand specific questions. Understanding the kinds of supervised learning techniques that can be employed, such as Bayesian theory, provides food for thought on how businesses can rethink content for machine training strategies.

 

Bayesian theory works backwards from the question that had been asked by guesstimating the context of why the consumer asked the question in the first place. It creates a common sense knowledge that a machine needs to have to answer the question in a relevant way. Just understanding this concept alone is powerful in helping businesses assess the need to build better content ontologies that could add more precision and efficiency in the learning strategies adopted by machines. 

 

As the saying goes, there is always two sides to a coin. And in the case of AI, machine learning statistical techniques must be complemented by well-managed content training strategies to ensure robust implementation.


RE•WORK
RE•WORK

RE•WORK成立于2013年,宗旨为促进国际人工智慧及相关之研究,发展,应用及交流。迄今为止,RE•WORK在世界各地已创办超过50场人工智能高峰会,包括新加坡,香港,美国,伦敦,加拿大等等。

产业语音识别自然语言处理机器学习
暂无评论
暂无评论~