Artificial intelligence has been the hot topic since ChatGPT rolled out nearly three years ago, and logistics companies have been finding new ways to implement it across their operations. These companies know there is value in AI, but many struggle to find a place to start.
The AI experts say to start small.
Start with the low-hanging fruit – the less complex problems that can be solved with simple things like ChatGPT or by building a custom solution on OpenAI’s application programming interface platform, said Toki Sherbakov, head of solutions architecture at ChatGPT’s parent company OpenAI.
“I think people think AI should be used for the most complex, highest value problem. We don't encourage you to start there. Maybe it's valuable, but it's going to be a little harder to get success there,” Sherbakov said during a session at Samsara’s recent Beyond user conference in San Diego.
“Don't over complicate things early. Don't boil the ocean with AI. Start on the lower complexity things, put it in the hands of your employees to build that AI literacy and enable everyone to know how to use it for their day to day, and then expand from there.”
He said new ideas will spark from employees as they build their AI literacy.
Enabling an AI workforce
Samsara hosts a weekly learning and enablement series where it invites its employees to come and set up AI tools and shares best practices and tips. It’s also a forum for teams to share their stories about how they’ve used AI and what has worked for them, said Evan Welbourne, vice president of machine learning and engineering at Samsara.
It's all about getting started, said Zach Merritt, director of data science and AI at Samsara.
“Get people trying it out,” he said. “Once they do, there's this sort of snowball effect of, ‘Wow, this is awesome. Could it do this? Could it do this?”
Samsara also has what it calls AI champions – employees who aren’t necessarily super technical individuals but who are really curious and like to try new things – in each department to give other employees ideas about how they could possibly use the technology.
“The champions network, that idea is also something that is very applicable across organizations,” said Sean McGee, Samsara’s head of product management.
AI in the passenger seat
Don’t throw AI at everything, Sherbakov said, adding that people think AI can solve all their problems, but that’s not the case. Sometimes machine learning will solve the problem.
“You have to just know where you're trying to get more of a predictive answer, a more deterministic answer. That's where machine learning can be very good,” he said. “If you're okay with a little bit of stochasticity and generative AI elements, then you can go there. I think if you use them thoughtfully together, it will give you the biggest impact.”
Welbourne said one tactical reason Samsara uses classical machine learning is because many of the best generative AI models won't fit on a dash camera because they use too much thermal energy required to run a model like that.
The company uses what he called classical AI, like deep learning neural networks, that it can scale down and run on the edge to do things like detect when a driver is drowsy or is using a cellphone. Generative AI, he said, is factored in later to do things like processing what was detected on the edge to gain additional context.
“We use both really in many of the solutions increasingly in the last year to see a lot of this combination of classical AI providing sort of point solutions and then generative AI tying it together to complete the experience,” Welbourne said.
