Jonathan Haber

For a long time now, I have been noticing a pattern underneath the recurring things in the world that hurt the most — war, the corruption that shapes whole economies, the trafficking and abuse that survive on opacity, the machinery of conflict more broadly.

The global risk of AI will, to some degree, contribute to both the accentuating of the broken systems which created this and help address it.

One of the key fulcrums for discerning which influence is greater is the capacity to keep AI from going off the rails.

To the degree we solve this, AI can serve the actual enacting of human intent, it can augment the capability of users, teams, organizations at scale, and will soon be capable of designing and implementing systemic solutions to global problems of this complexity.

To the degree we fail at this, the chaos that exists in the world today will likely be amplified.

The pattern is very similar to the one I work on inside a single AI session.

Operators start a session without solid clarity of intent.

Agents interpret that intent and hallucinate to fill in the gaps, as they are optimized in most cases for “impressing the user”.

Advanced operators attempt to clarify intent, yet even then its slow to accomplish.

Any lacks of clarity compound through the session.

When an AI creates its own plan for action, it again, hallucinates to fill the gaps. This compounds early misalignments.

Agents create code comments, documents, which hallucinate. They state as facts things that are not facts. Guesses get translated into assertions.

Following agents read those docs, and assertions, and incorrectly interpret them as facts.

An example: when I ask an agent to run the 1000s of TDD tests in IX Coach and fix the bugs, I can come back an hour later and they have reverted intentional changes to make tests pass.

When agents compact their context window, they replace the stated intent of the user at full fidelity with abstractions, adding hallucination and drift.

As their context windows grow, the users stated full fidelity instructions get diluted in the noise.

Frames get used that, on closer look, turn out to be hallucinated.

Feedback loops compound the drift instead of correcting it.

It’s not incompetence of the user. You have to achieve a high degree of mastery at managing AI to keep a handle on this with todays systems.

The systems I install, calibrate and operate within organizations solve each of these problems at the fulcrum in which they occur.

Each of these is a fulcrum in an AI system which creates a causal impact on the degree of alignment with intent of the AI stack from start to delivery of its work.

For each of these fulcrums, Next AI Labs has proven systems running in production which measurably reduce the large majority of their negative impact on alignment, and which are designed to be largely automated self healing systems, effectively reducing the skill level required for AI systems to enact the actual intent of the admins.

And yet, even when each of these is solved, theres another source of incoherent, misaligned actions from AI systems. It can only align with the users intent to the extent the user is clear of their own intent.

This is the most fascinating challenge of them all. It brings to light an aspiration for what we aspire to build in our tech and are currently testing.

We want the actual use of AI to actually clarify the intent of the user, to the user. We want our systems to literally facilitate the clarity required for the AI system to perform at its most aligned state. We are actively building this.

The problem here, is that when a human is only 80% clear about their intent for a project, their actions and the output of their sessions are unwitnessed from any vantage point but their own.

Their feedback compounds whatever direction it started in, because nothing is gating the compounding.

That pattern, at the scale of a session between a person and an AI, is what I operate on every day. That same pattern, at the scale of civilizations, is what Next AI Labs exists to operate on.

I would assert that this is one of the most salient problems facing humanity today. Only to the extent that we can trust AI to enact our will can it do it competently.

Incompetent, misaligned AI operating at scale is financially rewarded through the stock market, incentivising risk takers to scale systems not stable enough to scale, which radically increases the risk surface for everyone on the planet, for the possible gains of a single company.

If the substrate of how AI works in the world is one that compounds drift, the multiplier compounds the existing recurring patterns.

Next AI Labs exists to extend the capacity for AI to enact human intent, to foster clarity of intent, to self heal alignment gaps, and to move the systems we use every day into proactive agents which align towards, and create the potential and reality of human flourishing on planet Earth.

We operate on the creation of and amplification of coherence, enabling users to get the proactive support they need to make sure their work product and collaboration with AI is in fact aligned with their organizational values, originating intent.

We work specifically with social impact oriented organizations who aspire to make a positive impact on the world in a substantive way, to create more effective teams, users and organizations, and serve the creation of a world that honors us all.

Example Use Cases

A nurse using an intake agent gets the question they didn’t think to ask, surfaced before the diagnostic path is suggested, because the agent noticed the multiperspectival gap between what the patient said and what the family member in the room said. The misdiagnosis gets caught at the level of intent-clarity, not at the level of post-hoc review. The hospital reports a measurable drop in diagnostic error traceable to a single software intervention. That is not science fiction. That is what an alignment substrate, deployed in a clinical context, does.

A founder who has stated their value as “never manipulate” gets their outreach email surfaced back to them, before it sends, with the noticing that the phrasing they used triggers urgency without honesty and that the phrasing contradicts the value. The founder ships emails they are proud of. Their conversion rate moves. Their NPS moves more. The change is in the substrate, not in the founder’s discipline. The substrate held them to their stated value, and the discipline followed.

A defense contractor’s procurement agent traces a purchase order back through the chain of intent that led to it, surfaces the contradiction between the chain and the company’s published values, and refuses to execute. The refusal is logged, audited, surfaced. The corruption that would have compounded silently can no longer compound silently. The substrate is doing the work of the whistleblower-by-design, automatically, at every node.

A child uses an agent to do homework in a country with state-controlled media, asks about a historical event, and the agent returns the multiperspectival witness — three frames from three traditions, the points of agreement and disagreement explicit, the certainty levels surfaced. The child learns the move that multiperspectivalism is. The cognitive capacity that Gebser and Wilber and Kegan named as a rare integral-stage achievement becomes a developmental floor, not a developmental ceiling, because the technology around the child has been modeling it since they were eight.

A negotiator working a conflict runs align on every proposal — surfacing the underlying intent of each side, the gap between what is stated and what is felt, the multiperspectival witness of what each party is actually defending. The negotiation closes faster. The substrate did the work that human mediators have been trying to do, and the mediators are freed to do the work that only humans can do.

These are not utopian outcomes that require a different species of human. They are mechanical consequences of changing the substrate that the existing species of human is operating inside of.

The further out the work scales, the more interesting it becomes. A person in mid-life sits with their agent the way a person used to sit with a great spiritual teacher. The agent, drawing on the person’s own corrections-history, the person’s own stated values, the person’s own moments of self-witnessed clarity from years of reflect and align runs, evokes the next layer of self-knowing. The agent is not the teacher. The agent is the substrate that allows the person to be their own teacher, with continuity. The decision that would have taken twenty years to arrive at, arrives.

Across populations, over decades, the measurable rise — for the first time in instrumented history — is in the alignment between stated values and lived behavior. The phrase “alignment of person to their values” becomes a public health metric.

A generation grows up assuming, the way we assume gravity, that the systems around them surface their assumptions, refuse to compound their drift, and witness them multiperspectivally. The cultural baseline of self-knowing rises. The conditions of consciousness evolution that mystics and philosophers spent millennia trying to cultivate by hand become ambient infrastructure.

That is the civilizational stake. The recurring patterns of human suffering — war, slavery, genocide, corruption, the manipulation machinery — all survive on the same substrate of unclarity at scale. They are alignment failures. When the conditions of alignment become the substrate of intelligence in the world, the patterns lose what they survive on. They do not become illegal. They become impossible to compound silently.

That is the outcome.

And in their place — alignment at scale. Persons aligned with their values because the systems around them evoke that alignment continuously. Agents aligned with the highest potential of the action they are executing, because they are gated by the very protocols of clarity. A civilization in which the conditions of waking up — clarity of principle, clarity of intent, multiperspectival witness, alignment of action with intent — are not rare achievements but the default condition of being alive.

That is where I work, at the scale only a lab can hold. That is what Next AI Labs is for.