SAN FRANCISCO -- Recently, Facebook, Google, Amazon, and Microsoft and other big tech vendors have received criticism for relying on contractors to parse conversations and recordings made by users to help train the vendors’ AI systems.
By not explicitly telling users their “private” communications and AI assistant requests would be read or listened to, the tech giants violated their users’ privacy, critics say.
While the tech giants should have been more open as to how they would use user data, having a human involved not uncommon when training or using AI systems. In fact, pairing AI and humans is necessary for an accurate, intelligent AI system.
Full automation isn’t always best
“The only thing that’s being truly automated right now is the RPA,” said Ryan Welsh, CEO of AI vendor Kyndi, during a panel discussion on the future of AI at the AI Summit here.
Yet, beyond RPA, which is mostly used to automate repetitive tasks, fully automated systems don’t necessarily work well.
To illustrate the problems of fully automating a system, Welsh recalled working with a particular pharmaceutical company. The company, he said, would analyze large amounts of data to determine if anything went wrong in their manufacturing process.
The pharma firm wanted to automate that process, but when it provided Kyndi with a data set, Kyndi could not produce accurate results.
After analyzing what went wrong, Kyndi determined that the company’s employees who worked on the project only agreed with each other 66%of the time.
“So, if you try to automate a process that even your humans don’t agree on certain things you’d be better off giving those individuals AI to increase their productivity, and ideally, improve to process as opposed to trying to automate across an entire group,” Welsh said.
Human employees don’t agree on many processes that human employees don’t agree on so trying to automate all those processes won’t always give high-accuracy results, he said.
By pairing AI and humans, and using automation in specific areas rather than trying to automate an entire workflow, organizations can achieve faster results while retaining accuracy levels.
“All the AI that we see today is narrow AI,” said Asha Samal, senior director of AI at digital transformation vendor Publicis Sapient, during the panel discussion.
With narrow AI, AI and humans have to work together, Samal said. “It still requires that human-in-the-loop,” she said.
“In its current state, AI can learn, but it can’t reason,” Samal said.
General AI, however, could reason. It’s expected “to be able to use judgment in difficult situations, dip into prior knowledge” to help make decisions, Samal said.
Yet, she said, even with general AI, human will still have jobs and make decisions. Those jobs may be different than the jobs available today, but humans will always be relevant.
The 2019 AI Summit was held Sept. 25-26 at the Palace of Fine Arts.