connect ideas and people, identify [the] situation, and see what makes sense going forward,” Kerr added.
On the other hand, as one San Diego tech entrepreneur and former Google researcher (who would only speak anonymously) told me, “Google is absolutely looking for properties in San Diego. Of course, they’re always looking for innovative technologies. But San Diego is one of the tech hubs with the kind of startup ecosystem that can generate those types of companies.” As a tech startup hub, he added, San Diego is more or less comparable to Seattle, Boulder-Denver, and Austin, TX.
Some interesting connections already are taking shape, as San Diego’s expertise in genomics and other fields of “big biology” come together with Google’s expertise in “big data.” Exhibit No. 1 is last year’s appointment of Google’s Jeff Huber as the CEO of Grail, the diagnostic company spun out by San Diego-based Illumina (NASDAQ: [[ticker:ILMN), with the audacious goal of developing a blood test for cancer.
In his public presentation, Cerf put San Diego’s resident expertise in genomics, proteomics, and metabolomics at the top of his list of regional strengths.
He sees an intriguing mix of life sciences and technology R&D coming together, and he said the presence of Illumina in San Diego and Thermo Fisher Scientific’s (NYSE: [[ticker:TMO]]) Life Sciences Solutions in nearby Carlsbad, presents some huge opportunities for applying machine learning to what he called “biodata.”
On the other hand, Cerf said, San Diego is missing the kind of high-speed network infrastructure needed to facilitate such large-scale analytic efforts. Such a network, he said, would connect private companies to each other and to high-performance computing and data resources, including cloud-based services. The right IT infrastructure, Cerf added, would open huge opportunities for machine learning in other fields as well, such as cybersecurity.
With regard to the use of machine learning, Cerf said, “One of the things that is really important is if you’re going to teach a machine to do something, whether it is a discrimination task or functional task, like driving, you have to have really good quality data and it has to cover as many of the situations as you can possibly think of. Otherwise the system will not behave the way you expect it to. One possibility is to have a whole bunch of neural networks [available] that have [each] learned specialized things to do. Then you pick which one to apply, depending on the problems you need to solve.”
One of the recurring themes of his conversations with local entrepreneurs, investors, and startup founders, Cerf said, was that San Diego startups have to go to Silicon Valley to get money and engineers. “I remember thinking ‘Why is that?’ You’ve got good schools turning out good quality people. You’ve got companies you can raid engineers from. You have a number of venture capital firms. But apparently, it is still the case that there’s a lot more venture money in Silicon Valley than there is here. So we need to correct that.”
At another juncture, Cerf said that simply counting the number of startups is no way for a city to measure its success in creating a robust startup community.
“About 95 percent of all ventures fail,” Cerf said. “Every venture capital company knows that, and they hope that they will have five-percent success, and maybe even a unicorn, in order to cover the losses for companies that didn’t make it.”
“The real heart of successful business is innovation, and that comes out of research,” Cerf said. “That can be transferred in any number of ways. It’s not just by startups. Those transfers can take place into corporations like Qualcomm, which is a great big company, but it’s fully capable of absorbing new ideas and of generating new ones. So if we’re trying to measure how we’re doing in the innovation space, please don’t just count startups… Another really good metric would be how many startups are still alive after five years.”
Cerf also voiced strong concerns about what he called “autonomous software” during his presentation. “I am actually quite worried about software that decides to do stuff without you in the middle or any human being in the loop,” Cerf said. “If you think about the Internet of things, that’s exactly what we’re doing. We’re putting software into devices and letting the software decide what to do. In the first order, that is not necessarily a bad thing. For example, your thermostat tries to keep the temperature at what you want it to be.
“But you know, a self-driving car is a little bit more elaborate, and it’s running on its own. And what if it doesn’t do what you want it to do? What if there’s a bug that the people who wrote the software didn’t detect, and the car does something unexpected?
“So, I just worry about being too lax about software. The reason for this is that I used to make a living writing programs and in the last 80 years… we haven’t figured out how to write software that doesn’t have bugs. We don’t have very good tools for figuring that out either… So we have to make sure that we can fix the bugs, and that’s a challenge, because if you know that you need to load new software into a device, then the device needs to know that the software is coming from the right source, and what if it isn’t? What if it’s from someone who wants to put malware onto your devices? Now you’re in trouble. This is a non-trivial thing.”
Near the end of his talk, Cerf said he had invited entrepreneurs who participated in the sessions to submit a one-page summary about their startup idea, technology, or research. Kerr said she would compile the summaries and ultimately provide a curated list to Cerf “for possible presentation to Google or other interested parties.”
