Fred Anthony on Building AI That Learns From Experience
The technology executive and Surely Networks co-founder draws on hands-on engineering experience to explore AI that learns from evidence while remaining accountable to the people using it.
An experienced colleague does not need to hear the entire history of a project before deciding what to do next. They remember what mattered, recognize what has changed, and draw on earlier experience.
AI can answer difficult questions, yet an assistant may help resolve an issue today and require the same background again next week. The problem is bigger than forgetting a conversation. A system may remember an earlier conclusion without recognizing that new evidence makes it unreliable.
An Engineer at Heart
Fred has spent more than 25 years in software engineering and technology leadership. Even as his responsibilities expanded into executive roles, he remained involved in architecture and difficult technical decisions.
His career highlights include serving as Vice President of Technology at JoVE, where his work connected AI and educational technology. He helped modernize platforms and introduce AI into engineering workflows in an organization devoted to scientific education.
At Brain Scientific, Fred helped develop an AI-focused neurology platform that connected a disposable EEG headset with a remote amplifier and Neurology Hub. The work joined engineering with compliance and reinforced the importance of reliable signals and evidence when technology is used in health-related settings.
Startups demand quick feedback from real users without unnecessary complexity. Established organizations need to improve their systems without disrupting customers who depend on them. Fred approaches both by first understanding what the business needs and how people work, then choosing technology that fits.
He believes engineering leaders need to be close enough to the work to understand the tradeoffs while giving their teams room to challenge assumptions and take ownership.
From Memory to Adaptive Intelligence
As CTO and co-founder of Surely Networks, Fred is helping make AI more accessible by improving how it retains and uses context. People should not have to organize their work around an assistant’s limitations or repeatedly explain information it has already encountered.
He describes the architectural approach guiding this work as Evidence-Grounded Adaptive Intelligence (EGAI). It combines connected knowledge with feedback from experience so that AI can reconsider its understanding and make better-informed decisions. The point is not simply to store more information.
In this approach, original sources remain distinct from the system’s interpretations. A decision is linked to the circumstances behind it. An outcome provides evidence about whether the decision helped. A lesson captures what is worth carrying into future work, including the situations where it does not apply.
The architecture is informed by the way people connect experiences, but does not claim to recreate the human brain. Connections between knowledge can develop through use, while earlier conclusions remain open to correction. The emphasis is on adaptive behavior supported by evidence, not on a particular way to store memory.
When AI Questions What It Knows
Fred’s engineering work includes mechanisms for distinguishing decisions from lessons, tracking when knowledge has been replaced, and identifying gaps that need attention. Those capabilities give adaptive intelligence a practical foundation.
Consider an assistant that encounters the same difficulty several times. A perfect transcript of its attempts is not enough if it repeats an approach that failed. It needs a way to recognize the pattern and ask what information is missing.
In the knowledge-management workflows Fred has been developing, recurring issues can surface as questions. Answers can become guidance for subsequent work. The system also examines whether learned information is connected to the contexts where it can actually help. A lesson that cannot be found when needed has little value.
Fred also places importance on testing whether a new lesson improves outcomes. Adding knowledge should not automatically be treated as progress; it needs to be compared with the results of earlier work.
For him, this is the reflexive aspect of the underlying engineering: the system scrutinizes the understanding on which it relies, rather than treating its accumulated knowledge as permanently correct. EGAI uses that feedback to inform reasoning and action.
Learning Through Different Perspectives
Another part of Fred’s approach comes from earlier work designing AI advocates and mediation systems. Instead of treating one interpretation as complete, advocates contribute perspectives that the system considers alongside the available evidence.
Education illustrates the idea. A teacher needs to know how a student is performing, but grades alone do not explain how that student approaches a problem. A learner may repeatedly make the same mathematical mistake because of a misconception that another explanation could resolve.
The teacher’s advocate brings the learning objective into the discussion. The student’s advocate adds context about earlier difficulties and explanations that have helped. A mediator evaluates those perspectives against the evidence before recommending an approach.
Neither perspective automatically becomes a fact. The system must be able to challenge a teacher’s assumptions or a student’s interpretation, and acknowledge when it needs more information. That is especially important when biases shape how someone’s abilities are understood.
What happens after the recommendation matters too. If the student makes progress, that outcome adds evidence about what worked in the particular situation. If not, it is a reason to reconsider. In EGAI, advocacy helps shape reasoning and learning, rather than serving only as a final check on an answer.
Learning With Boundaries
Fred believes useful AI should also be proactive. A system that notices a missing piece of information can bring a timely question forward. A change to the evidence behind a decision can prompt a recommendation to revisit the plans that depend on it.
Some work can take place automatically within permissions established by users. Other steps require human approval. Learning more does not mean earning the right to change those boundaries.
That separation is visible in the architecture work Fred has developed: learned guidance is treated differently from governing instructions, and the reasons behind changes to knowledge can be preserved. The same controls help keep a useful connection from becoming unauthorized access to sensitive information.
Curiosity Beyond Today’s Computing
Fred continues to work directly with AI systems and computing hardware, testing how technology performs rather than relying only on its promises. His curiosity also extends to quantum mechanics and the nature of space and time.
He follows the potential intersection of quantum computing and AI, including whether future algorithms can offer new approaches to difficult computational problems. Practical quantum advantage for mainstream AI remains unproven, but the research interests him for the possibilities it raises.
Making AI Useful
A convincing demonstration is not enough. AI becomes dependable when it can learn from experience without losing track of the evidence, recognize the limits of its understanding, and respond appropriately as circumstances change.
For Fred, Evidence-Grounded Adaptive Intelligence brings those principles together. It is an architectural direction grounded in systems he has built and continues to develop, not a claim that every possible capability is complete.
The goal is not an assistant that remembers everything or takes over every decision. It is one that becomes more helpful through experience and supports people in making better-informed choices.
More from Tech

PlayStation Cancelled Kojima's Next Game in June and Xbox Has Picked It Up
Physint moves to Microsoft after three decades of work between Hideo Kojima and Sony, and the new deal extends into film and television.

Fewer Than One in Ten Enterprise Applications Is Fully Observable, and Agents Are Now Writing the Code
Missing visibility used to slow human teams down, but agents now deploy at machine speed into estates that nobody can fully see, which turns a cost…

The Cost of Software Is No Longer the Licence, It Is the Number of Places People Have to Look
Every switch between an app, a device and a workflow asks a person to reorient, repeat themselves or make another small decision, and the bill for…