The first studies around productivity with AI are arriving. A working paper by the National Bureau of Economic Research with Stanford and MIT found that access to generative AI can increase workers’ productivity by 14% on average, as measured by the amount of customer issues the agents could resolve per hour.
The study found that the productivity increase disproportionately benefited workers with less skill and experience. The AI tools helped fill the experience gap. Workers using AI with two months tenure performed as well as workers with six months tenure who didn’t use AI.
As a result, the study found that high-skill workers don’t have as much to gain from using AI assistance since the AI recommendations are essentially imitating the knowledge the high-skill workers already possess.
Google announced that its Bard chatbot down can now help with programming. It can now generate, debug, and explain code in 20 programming languages. The company has also moved the engineering team responsible for AI chips into Google Cloud. They also combined their two AI research labs, DeepMind and Google Brain, as they look to accelerate their AI progress.
Google also has announced Cloud Security AI Workbench, a cybersecurity suite powered by a specialized “security” AI language model called Sec-PaLM. Cloud Security AI Workbench spans a range of new AI-powered tools, like Mandiant’s Threat Intelligence AI, which will leverage Sec-PaLM to find, summarize and act on security threats. Finally, the company also announced that they are opening its security products to integrations from competitors.
Bing, too, with more upgrades — support for LaTeX markup, a reduction in end-of-conversation triggers, and news grounding, as well as fixing problems with the chatbot triggering duplicate searches and reducing errors that would lead to the chatbot not giving a response.
The team at VPN Mentor sent over some data on bias in AI-generated images – yes, it’s what you think. Examples include DALL-E results for nurses returning predominately women, CEO mainly returning men, or basketball players returning primarily men.
Finally, with the concern over the coming AI Spam flood, I wanted to highlight some reporting by Vice, who have found the first obvious tell for those who are lazy about using it – leaving in the phrases “as an AI language model” and “I cannot generate inappropriate content.” Experts anticipate more sophisticated, personalized, and targeted spam generated with AI.
The headline data on productivity is validation of impression rather than revelation. You can give less skilled employees access to information faster. This continues to support the assistant rather than the replacement model.
When you note the apparent tell of the lousy use of AI, cutting and pasting AI content, like those not even editing the results, is lazy and shows the best applications. It should accelerate people rather than replace them. Which leads to improvements – this space iterates very fast.