English | EN 
“Have we already seen it all?” Sally Epstein began playfully, looking across the room. “Or is the show just getting started?” The audience voted 99% optimistically: the best was still to come. “Good,” she smiled. “Because that’s the only right answer.”
Epstein, Chief Innovation Officer at Cambridge Consultants (part of the Capgemini family), lives at the bleeding edge where AI meets deep technology — that realm of inventions so complex that even engineers whisper “let’s see if physics agrees.” Her world includes quantum computing, biotechnology, advanced robotics, and AI systems that teach themselves to think differently from humans. “Deep tech,” she explained, “isn’t about shiny apps or convenience gadgets. It’s about projects with high scientific uncertainty but huge potential impact. They take longer, they’re harder — but when they work, they change everything.”
When Robots Meet Humans (and It Gets Awkward)
To illustrate, Epstein first took the audience back six years to an experiment that mixed social science with robotics. “We built a team half of behavioral scientists and half of technologists,” she said, “because if you let computer scientists decide how a machine interacts with humans, they make dumb decisions.”
The result: an attempt at robotic therapy. Volunteers received emotional counseling from a humanoid robot. “And they absolutely hated it,” she laughed. “It turns out, not every human wants to cry in front of a metal face.” But not all experiments ended that way. Epstein’s team also created a project for dementia patients who could no longer express
their feelings. Using AI to read non-verbal cues — facial tension, body movement, micro expressions — the system selected personalized music playlists to lift moods. “We’re getting beautiful feedback from families,” she said. “It shows that even when AI feels creepy, it can still bring joy.”
Storing the Entirety of Wikipedia in a Test Tube
Next came a favorite project that sounded straight out of science fiction. “We were approached by a U.S. start-up that wanted to print data into DNA,” she said, pausing as the audience gasped. “At first, I thought they were mad. But DNA is nature’s ultimate hard drive — incredibly dense, stable for centuries, and doesn’t need refrigeration. You can literally store the entire Wikipedia in a vial the size of your pinky.”
She showed a photo of a colleague holding such a vial. “All of Wikipedia,” she repeated. “In there. Boom.”
When moderator Michiel later asked if she could put that DNA into a pill and “wake up knowing everything,” Epstein laughed. “Not yet,” she said. “It’s easy to write to DNA, but painfully hard to read it back. Your stomach isn’t quite up to it.”
AI at War — and Peace
Moving from biology to defense, Epstein described how her team applied AI to cyber resilient military vehicles. “Like every organization, the Ministry of Defense is undergoing digital transformation,” she explained. “Unfortunately, when you digitize tanks, you also make them hackable.”
In one project, vehicles under attack experienced GPS spoofing — enemies fed false coordinates to mislead them. The AI had to decide what to do in real time. “The safest thing would be to turn everything off and stop. But on the battlefield, stopping is not an option.”
So, the AI learned to reboot, isolate the fault, and continue. The goal wasn’t the perfect drive— it was the safest mission outcome. “We didn’t train it to win,” she said. “We trained it to survive.”
Her voice carried pride. “As far as we know, only two organizations in the world have trained agents at this complexity — us and DeepMind.”
Humanoid Robots: The Return of the Two-Legged Dream
“Why,” she asked next, “are we suddenly obsessed with humanoid robots again?” The answer was practical: our world is built for two legs and two arms. Door handles, stairs, shelves — everything assumes a human form. “So, it makes sense to design robots that can navigate our spaces,” she said. “Even if, honestly, no one believes they need to look like us.”
The breakthrough isn’t mechanical — it’s cognitive. “The challenge isn’t the robot’s body,” she explained. “It’s teaching it to move like a human — fluidly, safely, intelligently.”
Traditional programming made that impossible. But the arrival of visual-language-action models (VLAMs) — cousins of large language models like ChatGPT — changed everything. “Before, you had to manually program every move. Now you just say, ‘Hey robot, pick up the green square,’ and it learns what that means.”
She showed a clip of her colleague Ali interacting with a robot.
“Can you move this cup over there?” he asked. The robot hesitated: “Hmm. There are two cups. Please point.”
“Do you want me to move the first cup to the left?”
“Yes, please,” said Ali, weary.
“Poor Ali,” Epstein smiled. “He’s been stuck in that room for weeks. We’ll let him out soon.”
The audience laughed, but her point was serious: “We’re making progress, but our tolerance for robotic failure is incredibly low. When a human fumbles a cup, we sigh. When a robot does, we lose faith. We expect perfection from our machines.”
AI’s Role in the Lab: Designing Life
Then came the highlight — a story that truly lived up to her title “You Ain’t Seen Nothing Yet.”
A few years ago, Japanese scientists discovered a natural enzyme in a garbage dump that could eat PET plastic — a potential solution to the global waste crisis. The only problem: “It took a month to digest a single bottle cap,” Epstein said. “And only under perfect lab conditions. Not ideal for cleaning up the oceans.”
So, her team asked a radical question: could AI design a better enzyme?
Proteins, she explained, are just long sequences of amino acids — biological sentences written in an alphabet of A, T, G, and C. “So, we thought: if large language models can write poetry, why not proteins?”
They trained an AI model on vast protein databases. After a year, it produced redesigned enzymes that were five to seven times more effective than nature’s original. “Our biologists were stunned,” she said. “The AI made changes in 60 sites across the protein — places human experts had said were irrelevant. It just ignored us, and it was right.”
She showed slides comparing the glowing green of human-designed proteins versus AI-designed ones. “The AI ones are naughty,” she said, laughing. “They just refuse to follow the rules. But they shine a lot brighter.”
Of Dishwashers, Robots, and Cat Litter
In the Q&A, Epstein fielded questions ranging from the whimsical to the practical. When asked how soon a household robot could unload a dishwasher, she replied: “Honestly? Not soon enough. Some people name their robot vacuums — ours is called Dina the Warrior —
but she once picked up a knife, so I’m not sure I’m ready to trust her in my kitchen.”
She added that humanoid robots are far more useful in industrial settings: logistics, warehouses, nuclear facilities. “And oddly enough,” she said, “people love having them in the office. They just want to take selfies with them.”
The anecdote about airport robots drew laughter: “At Schiphol, people jump in front of the robotic wheelchairs just to see if they stop. They tease them! We even had to train theme park robots to deal with pranksters who block their way.” The moral, she said, was that “the hardest part of robotics isn’t mechanics — it’s humans.”
Conclusion: Beyond the Horizon
Epstein closed with a challenge. “Let’s look beyond agents and chatbots,” she urged. “Let’s imagine AI redesigning biology, reshaping materials, protecting lives, and yes, maybe even understanding us one day.” Her smile turned conspiratorial. “We’ve only scratched the surface. So if anyone tells you AI has peaked…” — she paused — “just tell them: you ain’t seen nothing yet.”
Get your copy of the Autopilot Yes/No Report.
Please note – This report was created by almost exclusively using available AI-tools except for minor editorial tweaks and some limited lay-out changes.