Is there too much hype around AI?
Although AI technology is being used very successfully in those projects, Diedrich warns against seeing it as the magic bullet that the media are currently making it out to be. “AI is effective, no doubt, but it is being overhyped right now,” says the head of Business Development. “Every user needs to ask himself: ‘What is my actual problem? What is really behind it? Can I increase the benefit to myself and my customers through AI?’ And sometimes the answer might be: ‘No, it’s not worth it at the moment.”
Strong words from an agency, you might think. But this sort of honesty is very important to hmmh. “We do rely on AI where it really makes sense, and is economical. But we don’t engage in hype, we deliver solid performance,” says Diedrich firmly.
Data, data, data
One place that delivers solid results at hmmh is the MAD unit. The name stands for Mobile Apps & Devices. Within the agency, a unit is a department that deals with all aspects of a customer’s requirements relating to a specific issue. MAD is located on the 15th floor of Bremen’s Weser Tower, one of the top floors of the eight occupied by hmmh in the impressive building on the edge of Bremen’s Überseestadt district. MAD develops concepts and technologies. Here, programmers work side by side with creatives on apps, augmented and virtual reality, and above all on devices for the Internet of Things. They also process machine data for new applications.
This data is crucial when it comes to using AI. “There has to be an ecosystem around AI, otherwise the project won’t get anywhere,” says Diedrich up on the 15th floor. “The first questions that companies moving towards AI need to ask themselves are: ‘What data do I have? Am I collecting machine data in the production department or financial data in controlling? Can marketing provide some figures? How can I gather more data within the company and process it in a way that makes it compatible with AI applications?’”
Data can take all sorts of forms, such as photos, audio recordings, videos, sensor readings, or tables of user behaviour from the website. “Many companies do not have sufficient base data, and if they do, it is not structured,” Diedrich explains. That means the AI will not be able to make use of it.
To structure and catalogue – that’s half the battle
To find out why properly categorised and processed data is of such importance, we need to move from the 15th to the 9th floor. Instead of small, team-based offices this has an open-plan area which covers almost the entire floor, with individual workplaces crammed tightly together. “When you’re using an AI, 60 to 70 per cent of the work is data processing. And that is one of the things we are doing here,” says Diedrich as he looks round.
Any AI is only as good as the data it runs on. At the start of every project, the existing data must be prepared, processed and adapted. Despite all the automation, that work can only be done manually. “To train an AI to recognise images we must first of all define what those images show in the training data,” Diedrich explains. This work is a necessary initial investment.
The 22-storey Weser Tower: an impressive Bremen landmark