It is time to stop anthropomorphizing AI and start treating it for what it actually is: a mathematical node in a larger collaborative group. When we obsess over whether a machine can pass the Turing Test, we fall into the trap of valuing mimicry over utility. To truly harness this technology, we must move from a framework of “human versus machine” to one of “AI in the group,” where success is measured by collective performance rather than believable imitation.

The Evolution of Alien Interactions

The term “AI” has become a catch-all, but the field has existed under many names since the 1950s. In many ways, it may be better to consider AI as “Alien Interactions.” These systems do not think like us. They process information through architectures fundamentally different from human cognition.

Over the decades, we have seen many flavors emerge: Logical Reasoning and Problem-Solving Algorithms. Expert Systems. Statistical Inferences and Reasoning. Decision Support Systems. Cognitive Simulation. Natural Language Processing. Machine Learning. Neural Networks. And today, Large Language Models. Each represents a different approach to computational problem-solving, each with distinct capabilities and limitations.

Understanding this history matters because it reminds us that “AI” is not a single thing. When we treat all AI as if it were the same, we make poor decisions about deployment, governance, and trust.

The Wrong Test for the Wrong Goal

The original Turing Test, proposed in 1950, involved Computer A and Person B attempting to convince Person C that they were human. It was a test of deception and mimicry. Yet what if a computer “fooling us” that it was similar to a human, was in fact the wrong test entirely?

Instead of asking whether Computer A can convince Human C that it is human, we should be asking: How well was individual Human B, or a group of humans, doing at a particular activity before Computer A arrived?

This includes making better decisions. Or, perhaps more realistically in complex environments, making less bad decisions less often.

This reframing changes everything about how we evaluate, deploy, and govern AI systems. A customer service chatbot that perfectly mimics human conversation but frustrates customers and fails to resolve issues is a failure, regardless of how “human” it sounds.

My friend and colleague Vint Cerf, one of the Internet's co-originators, proposed an inversion of the Turing Test in 2018. In what he called “Turing Test 2,” a computer program interacts with both a human and another computer, attempting to distinguish between them.

Cerf's inversion addresses a real security challenge: malicious bots emulating humans to spread inauthentic information, launch phishing attacks, create millions of fake social media accounts, and pollute crowdsourcing systems. In 2017 I witnessed an early precursor of what, now in the 2020s, has become commonplace with bots generating inauthentic human-like behaviors. Yet even this formulation still centers on distinguishing human from machine.

The deeper question remains: What are we trying to accomplish, and does the AI help us accomplish it better?

Large Language Models, a subset of Generative AI, are essentially pattern-matching engines. They are entirely dependent on past training data. They do not “know” things in the way humans do. They predict the next likely piece of information based on what they have seen before. This is not a flaw. It is simply what they are. Understanding this helps us use them appropriately and avoid misplaced expectations.

The Architecture of Trust

To work effectively with these systems, we must understand the nature of trust itself. Trust is the willingness to be vulnerable to the actions of an actor not directly controlled by you. This applies whether that actor is a colleague, an organization, or a computational system.

Research on human cognition shows that we humans tend to trust an actor if we perceive three specific antecedents:

  • Perceived Benevolence: The belief that the actor wants to do good by us, that their intentions align with our interests.
  • Perceived Competence: The belief that the actor has the skills and capabilities to perform the task effectively.
  • Perceived Integrity: The belief that the actor adheres to a set of principles and will not readily change their stance based on convenience or pressure.

When we apply these antecedents to current Large Language Models and chatbots, the results are concerning.

Benevolence is indeterminate. These systems do not have intentions in any meaningful sense. They optimize for patterns in training data and parameters set by their creators. Whether that optimization serves our interests depends entirely on how the system was designed and deployed.

Competence is questionable. LLMs, by themselves, are not fact-checking systems. They can spread inauthentic information with the same confidence they display when providing accurate information. They have no internal mechanism to distinguish truth from plausible-sounding falsehood.

Integrity is largely absent. These systems will change their stance readily based on how a prompt is phrased. They have no consistent principles beyond the statistical patterns in their training data.

If we cannot reliably perceive benevolence, competence, or integrity in these systems, how can we trust them?

Yet before we conclude that AI systems are uniquely untrustworthy, we should consider other “obscured boxes” in our society. How much do we actually know about the decision-making processes in large organizations, government agencies, or corporate boards? The challenge of trust is not unique to AI. Remedying widening “trust gaps” represents both an important and fundamental challenge of complex societies where we must rely on actors and systems we cannot directly control or fully understand.

The Illusion of Perfect Control

Part of the challenge in building trustworthy human-AI collaborations is that humans want a governance illusion of perfect deterministic control over what AI does. For certain types of AI, rules-based systems and traditional algorithms, that is possible. You can specify exactly what the system will do in every circumstance.

However, for generative AI, the generative nature is a feature, not a bug. The ability to produce novel outputs, to synthesize information in unexpected ways, to identify patterns humans might miss, this is precisely what makes these systems valuable. This governance illusion of perfect deterministic control over such systems is a pipe dream.

Yet if we step back, we will realize that even without AI, we do not have perfect deterministic control over what humans do in the workplace. Sure, you can have guidelines, rules, and laws. But that does not mean a human will follow them. The illusion that we can ensure no human ever deviates from what was originally intended is exactly that: an illusion.

We have built entire systems of governance around this reality. We do not expect perfect compliance from humans. We expect accountability, learning from mistakes, continuous improvement, and adaptation to changing circumstances. We build in redundancy, oversight, and feedback loops.

Why, then, do we demand a different standard from AI systems working alongside humans?

The answer often comes down to fear of the unfamiliar. When a human makes a mistake, we understand the cognitive processes involved, even if we cannot predict them perfectly. When an AI system produces an unexpected output, it feels alien and uncontrollable. This emotional response is understandable. Yet it is not a sound basis for governance.

AI in the Group: A Different Framework

What we need is a different mental model entirely. Not “human versus AI.” Not even “human with AI” as if the AI is a tool entirely subordinate to human control.

Instead, we need to think of AI as one participant among many in a collaborative network. AI in the group. One node among others.

Not anthropomorphizing the machine (it is math, after all), but recognizing it as a participant in larger group collaboration. All participants, both human and AI, sometimes operate with some degree of autonomy and then circle back to share results, insights, or questions with the broader group.

This happens already in high-performing teams. A researcher goes off to analyze a dataset. An engineer prototypes a solution. A strategist consults with external experts. They work with varying degrees of autonomy, then bring their findings back to the group. The group does not exercise perfect deterministic control over what each member does during their autonomous work. The group benefits from multiple different viewpoints and the synthesis that happens when different approaches at solving a problem connect.

AI can function similarly. A diagnostic AI analyzes medical imaging while the physician reviews the patient's history and conducts the physical examination. A forecasting AI identifies patterns in economic data while human analysts consider geopolitical factors and qualitative information. Each operates with appropriate autonomy, then the results are integrated.

The question is not whether we can control exactly what the AI does at every step. The question is whether the group (humans and AI together) performs better than humans alone would have.

Rebuilding Trust for a Collaborative Future

Going forward, we will need to remedy and improve two things simultaneously: trust in our societies and trust in decision-making with humans and AI together.

Trust in our societies has been eroding for decades. People feel anxious about their ability to provide for themselves and their loved ones. They feel disconnected from institutions that seem opaque and unaccountable. This fragmentation serves no one well.

At the same time, we are deploying AI systems across domains that require trust: healthcare, education, national security, financial services, public infrastructure. If we cannot build trustworthy human-AI collaboration, we will either reject valuable capabilities or deploy systems that harm people.

These two challenges are connected.

The same factors that erode trust in human institutions (opacity, lack of accountability, perceived misalignment of interests) also erode trust in AI systems. The same approaches that can rebuild trust in societies (transparency, clear accountability, demonstrable competence, consistent principles) can also build trust in human-AI collaboration.

What This Means for Leaders

Leaders working across communities face a choice. We can continue treating AI as either a threat to be contained or a magic solution to be deployed everywhere. We can continue demanding perfect control while ignoring the reality that we do not have perfect control over human systems either. We can continue evaluating AI based on whether it seems human rather than whether it amplifies human skills and decisions.

Or, instead, we can build frameworks that treat AI as collaborative participants in work that matters. To take this path requires several practical steps.

First, we must evaluate AI systems based on performance outcomes, not mimicry of human behavior. Ask whether the AI helps people accomplish their goals more effectively. Ask whether it reduces errors, expands capabilities, or enables work that was not previously possible.

Second, we must abandon the illusion of perfect control. Build governance systems that acknowledge uncertainty and provide accountability without demanding deterministic predictability. Focus on resilience. Focus on outcomes, feedback loops, and continuous improvement rather than trying to specify every possible scenario in advance.

Third, we must invest in human capacity to work effectively with AI collaborators. This means understanding what AI can and cannot do. It means developing judgment about when to trust AI recommendations and when to override them. It means building teams that can integrate human and AI contributions effectively.

Fourth, we must design systems that treat AI as a participant in group collaboration rather than either a threat to human agency or a tool entirely subordinate to human control. This means creating appropriate autonomy for both humans and AI, with clear mechanisms for integration and accountability.

Fifth, we must recognize that security concerns remain valid even within this framework. We need to distinguish between legitimate AI participants in collaborative work and malicious bots designed to deceive, manipulate, or disrupt. But we should address this through security measures and authentication systems, not by demanding that all AI be perfectly controllable.

Sixth, we must build transparency about how AI systems work, what data they use, what they optimize for, and what their limitations are. This does not mean making every algorithm open source. It means providing enough information for people to make informed decisions about when to trust AI recommendations and when to apply their own judgment.

Seventh, and perhaps most importantly, we must establish clear accountability for outcomes. When human-AI collaboration produces good results, we should understand why and replicate it. When it produces poor results, we should understand why and fix it. This requires tracking outcomes, not just outputs, and building feedback loops that improve performance over time.

The Path Forward

The challenges we face as societies do not respect the boundaries between human and machine capabilities. Public health requires identifying disease patterns across populations while respecting individual privacy and autonomy. Economic opportunity requires matching skills to needs at scale while accounting for human dignity and community values. These challenges are too complex for humans alone and too nuanced for AI alone.

They require collaboration.

Seventy years after Turing proposed his test, we have the opportunity to move beyond mimicry as our measure of success.

We can build AI systems that amplify human skills and decisions rather than simply imitating human conversation. We can create governance frameworks that acknowledge reality rather than chasing illusions of perfect control. We can treat AI as collaborative participants in work that matters.

I submit the most important question is not whether AI can convince us it is human, because this is an illusion bordering on digital deception. Instead we should work together to build futures worth living in, with AI helping us achieve these futures. We should work on rebuilding trust in our societies while also building trust in human-AI collaboration. And most importantly, we should work to expand human agency and choice in an era of rapid technological change.

Ultimately, instead of the Turing Test, we should be testing how well any AI amplifies human skills and decisions both individually and collectively. Because this is the test that matters for both our current now and the future ahead.

Dr. David Bray is both Chair of the Accelerator and a Distinguished Fellow at the non-partisan Stimson Center as well as Principal and CEO at LeadDoAdapt Ventures, Inc. He previously served as a non-partisan Senior National Intelligence Service Executive, as Chief Information Officer of the Federal Communications Commission, and IT Chief for the Bioterrorism Preparedness and Response Program. Business Insider named him one of the top “24 Americans Changing the World” and he has received both the Joint Civilian Service Commendation Award and the National Intelligence Exceptional Achievement Medal. The U.S. Congress invited him to serve as an expert witness on AI in September 2025. He also advises corporate Boards and CEOs on navigating the convergence of AI, cybersecurity, and geopolitical risk.