I was secretly delighted when they would say that, because I had a standing order when that happened to refer them immediately to the White House, at which time the Chief of Staff would read them the riot act. A couple of times I understand the President himself got on the line to do the honors. I was never a huge fan of President Haden before the HRIA but I appreciated him after the fact, because he simply took no crap from anyone about how the HRIA was run. You opened up your research or you didn’t participate. And there was so much money involved that eventually everyone gave in.
Sharing data that way was not really the optimal way of doing things. If any of us at HHS could have changed it, we would have done things Manhattan Project style, where we sent all the researchers into the desert together until they came up with things we could use. But this set-up allowed the communal effort to have at least a thin veneer of free enterprise, and that was politically important, considering the administration.
And at the end of it all, it worked. The first neural networks came about because research on detecting brain activity by way of MRI and other external devices at Stanford was combined with physical deep brain stimulation research at the Cleveland Clinic by a scientist working at General Electric. If they hadn’t been able to see each other’s work, they all would have had to reinvent those particular wheels. This way the wheel only had to be invented once.
Heng Chang, neural network developer, General Electric:
Before Haden’s there was already a considerable amount of work being done in the field of directed brain imagery—using MRIs and other similar equipment to record and register when and how thoughts were being transmitted, and visualizing the brain as it responded to outside stimuli. At first for third-stage Haden’s patients, that’s how they communicated—sensors would be placed on their heads and scalps and we could very laboriously piece together their thoughts, sometimes just by running down the alphabet and having them think “yes” when we came to the letter they were thinking of. Spelling that way. Obviously that was a laborious process and not one that could be replicated for millions of Haden’s patients.
When GE started researching Haden’s, we got access to the HRIA database and as I was going through it I became intrigued at some experiments the Cleveland Clinic was doing with very sensitive antenna-like filaments they were developing to track incipient seizures in epilepsy patients, with the idea of then applying deep brain stimulation to arrest the seizures before they began. I thought to myself, wouldn’t it be great if the filaments could send as well as receive. You could use them to allow input from outside the body directly into the brain, and sent thoughts out the same way. Then I didn’t think about it again because I was working on another project entirely, and the Clinic’s work wasn’t on-point to that.
But my subconscious mind must have been still thinking about it because about a week later I came flying out of a dead sleep with the idea of the neural network. It was like it just downloaded directly into my brain. My reaction to it was so strong that I sat straight up in bed and actually shouted—not “Eureka!” but just a really loud gasp. This turned out not to be a great thing, because my cat was sleeping on my chest and was so surprised that she ended up digging into my skin before running off the bed. I got out of bed, swabbed the blood off my chest, and then drove to work in the middle of the night to start modeling the network I thought of. I didn’t want to go back to sleep. If I did that, I was pretty sure I was going to lose it entirely.
Ida Garza:
Chang’s idea was brilliant. Every scientist, on staff and off, told me so. So we knew this was a direction we needed to go, and quickly. What we didn’t know was how much it would cost. There’s an old saying: “Fast. Good. Cheap. Pick two.” Meaning that you’d never get all three at once. We picked fast and good. We didn’t assume it would be cheap, which as it turns out was a good thing.
Heng Chang:
They told me years later that by the time we got the first fully functional neural network into production, we had burned through something like a hundred billion dollars developing, testing and manufacturing it. That’s a literally inconceivable amount of money to me. Certainly I never saw any of it, other than my salary at GE. But I did get on the cover of Time magazine and was a finalist for Person of the Year, so that made my mother proud.
Irving Bennett:
So Chang and his team developed the neural network, but one drawback they had was in its testing. They could model the networks in supercomputers which could create environments that superficially resembled the human brain, and those models got them something like eighty five to ninety percent of the way there. But at the end of the day, if you want to find out whether they work, you have to put the networks into an actual brain.
And ultimately it has to be a human brain, for two reasons. One, because animals’ brains aren’t complex enough, and two, because an animal isn’t going to be able to talk to you about whether the network is functioning. There was also the catch that during the meningitis phase the Haden virus changed the structure and function of the brain so much that there was literally no useful analogue in the natural world for it. If you wanted to see how the networks worked in a Haden brain, you needed an actual Haden brain.
Naturally, this created a moral and ethical issue. These first neural networks were both highly experimental and highly invasive—the work papers Chang and his crew published described how the filaments of the network would need to penetrate and migrate through the brain matter, essentially turning the brain into a massive pincushion, without any guarantee that the invasion of this artificial neural network wouldn’t kill or debilitate these Haden’s patients even more than they were.
When I wrote up the stories on the work papers the families of Haden’s patients nearly rioted. They felt like their family members were about to be victimized a second time. President Haden had to cut short a trip to Indonesia to come back and deal with it. He was not happy about it, or with me. [New York Times Publisher] Bitsy Lapine called me into her office to let me know that the President had called and yelled at her about me for twenty minutes. Bitsy, bless her, eventually recited the First Amendment at him and hung up.
Heng Chang:
The media response to our initial set of work papers highlighted the problem that the public had in understanding how efficiently we could model the human brain, and the network’s interaction with it. We were very comfortable that the networks could be installed safely in nearly all cases. But there was always that tiny bit of risk that you can never get rid of, and so the press and the families focused on that. There was really no way we’d be able to get volunteers from the general Haden population after that.
Irving Bennett:
This is where the near-universal nature of Haden’s Syndrome came to the rescue. The first guinea pigs for the networks were eventually recruited from three distinct groups. The first was the very elderly—people who from an actuarial point of view only had a couple years of life in them even before they got sick with Haden’s. The second group was people with terminal illnesses who also got Haden’s—stage four cancer patients and patients with other advanced illnesses. These two groups had almost literally nothing to lose by volunteering to test these networks.