Who joins us at the table
As families gather for Thanksgiving, many will retell the same stories: how the business began, what the founder insisted on, what they would think of things today. Often there is an empty chair at the table, and the founder’s voice still has a way of joining the conversation. This year, technology offers a new way to bring that voice back. It is worth asking what we would be inviting to the table.
When the past starts talking
On September 15, 2026, at Grand Central Terminal in New York, Consuelo Vanderbilt Costin spoke with AI-powered holograms of four of her ancestors, including Cornelius Vanderbilt, who built one of America’s great business fortunes. The moment was part of a Fashion Week launch, so it was a brand event. It also showed something family enterprises will soon face. Generations separated by more than a century can now appear to hold a conversation.
Every family has stories about its founders. Their convictions become guiding principles, their decisions become precedents, and their words are repeated long after they are gone. Family members still ask what the founder would have done. Until recently, the answer depended on memories, documents, and interpretation. Now AI can generate a response in a familiar voice.
A founder who once lived through stories can appear to speak for themselves. When we give the past a voice, do we also give it authority?
What the research tells us
Two recent papers in Philosophy & Technology examine “deadbots”: AI systems that simulate a deceased person using their messages, recordings, and other personal data. Both are conceptual and ethical analyses. Neither studies family businesses or measures real effects on families, so we use them here as food for thought, not as evidence about family enterprises.
Who has a say. In 2024, Tomasz Hollanek and Katarzyna Nowaczyk-Basińska built three speculative scenarios. They distinguish three roles: the person whose data is used (the data donor), the person who holds that data after death (the data recipient), and the person meant to interact with the result (the service interactant). Their interests do not necessarily align.
In one scenario, a father in palliative care secretly creates a digital version of himself for his family. He does not ask his two adult children. After his death, one finds the experience draining and feels guilty about wanting to stop. The other prefers to grieve in his own way, yet keeps receiving messages. When they ask to deactivate it, the provider refuses, because their father prepaid a twenty-year subscription.
A loving intention can leave heirs with a relationship they never chose and cannot easily end. The authors therefore propose mutual consent. The wishes of the person recreated matter, and so does the freedom of those who will interact with the simulation. The authors also recommend dignified ways to retire these systems and clear transparency about what they are. Their recommendations are addressed to the companies that provide such services.
What the system can and cannot know. In April 2026, Atay Kozlovski, Edina Harbinja, and Roel Dobbe published a broader analysis of the uncertainty surrounding interactive deadbots. Two of their points matter for families.
First, technical safeguards help but do not settle the question. Restricting a system to verified material improves plausibility, but the authors note that such controls still do not make outputs deterministic. Even when an answer cites the record, we can know only that it is, at best, supported by the record, not that it is what the person would have said.
Second, many situations allow several answers consistent with a person’s values. When a system resolves such a case, the authors argue, the output reflects a choice made by the system, not the person. The authors also warn, in the context of grief, that answers implying “she would have chosen this” risk guiding users instead of informing them. In a family enterprise, the same dynamic could shape ownership decisions.
Together, the papers suggest that preserving a person’s identity and letting that identity keep influencing others are different things. There is a difference between asking what someone actually believed and asking an algorithm to imagine what they might believe today. Applying this to family enterprises is our own extension, not a finding of either paper.
When memory becomes authority
In family terms, the founder is the data donor. An owner, trustee, or family office may become the data recipient, holding the material. The next generation are the interactants, the people expected to engage with the result. The father’s problem in the scenario above could just as easily appear in a family holding company: who can say stop?
Some founder technologies already exist and are bounded. At the Walmart Museum in Bentonville, Arkansas, visitors can question a holographic Sam Walton. As publicly reported, it draws on his recorded words and archive material and does not generate new opinions. This is an archive that an institution or family can inspect and curate. Kozlovski and colleagues draw the same line, describing retrieval-based systems that surface what was actually said as different from tools that generate new statements in a person’s voice.
Now imagine a third-generation family debating whether to sell a division held for seventy years. Some owners see an opportunity to free resources. Others see a betrayal of the founder’s vision. Someone suggests consulting the AI founder.
Suppose it answers that the business should never be sold. Or suppose it gives a careful, ambiguous answer, and each side hears the confirmation it wanted. Even when everyone knows the response is generated, its apparent authenticity carries emotional weight. Those opposed to the sale can invoke it as the founder’s wish. Those in favor must justify their position to living relatives and to a digital version of their ancestor. No voting right has changed, yet the way authority works in the family has.
Technology helps, governance decides
Technical choices matter. A system limited to verified recordings and documents is more transparent than one that improvises. It can say “I don’t know” when the record is silent, restrict who has access, and be switched off. Families should insist on these features.
But they do not answer the questions that matter most, because those questions are about who holds authority:
- Who decides which memories, documents, and recordings feed the system, and who can change them?
- Should it explain past decisions, advise on new ones, or stay out of ownership discussions?
- Who can retire it, and what happens if relatives disagree?
- Can any family member decline to engage without penalty?
Software can enforce a decision, but it cannot make one. Kozlovski and colleagues note that some scholars propose moving from ownership toward stewardship, with duties of care, auditability, and sunset clauses. They also acknowledge that who should act as steward remains unresolved, since families, estates, communities, and institutions all have claims. Enterprising families already think in terms of stewardship, which makes this an old conversation in a new setting.
Families already govern values through constitutions and charters. A short digital legacy conversation could sit alongside them. The research cannot tell families what to decide, and the service-provider recommendations may never be adopted by any company. A family can still choose to talk about the questions the research raises. Four areas could be a starting point:
- The founder’s wishes. What, if anything, did the founder want preserved, and for what purposes? How would we know, and who speaks for them if we are unsure?
- Each person’s choice. How can each family member decide whether to engage with a digital founder, and change their mind later, without it affecting how they are seen in the family?
- Archive and speculation. How do we tell apart what the founder actually said and did from what a system infers? Does this difference matter in ownership discussions, and if so, how?
- Stewardship and endings. Who looks after such a tool, and how do we resolve disagreement about it? If it were ever retired, what would a respectful ending look like?
These questions have no single right answer, and families will answer them differently. Some may decide they want none of this. Others may welcome it with clear limits. The conversation itself is the point, because it surfaces what each generation values and expects before a decision has to be made under pressure. Some families may want to write their conclusions down alongside their constitution or charter, and others may prefer to revisit them from time to time.
Family enterprises rightly invest in preserving the knowledge, values, and experiences of earlier generations, and AI may offer remarkable new ways to do so. But a legacy is not simply something to preserve. It is also something each generation must interpret, challenge, and carry forward. The real test may not be whether we can keep a founder’s voice alive, but whether future generations remain free to disagree with it.
Questions for reflection
- If you could have one conversation with your family’s founder today, what would you want to understand about their thinking, rather than simply hear their advice?
- Where would you draw the line between preserving a founder’s wisdom and allowing that wisdom to influence decisions the founder never faced?
- Who should have the authority to create, shape, or retire a digital representation of a family member, especially when relatives disagree?
- How can families make sure that preserving the voices of previous generations does not come at the expense of developing the voices of those who follow?
Closing comments
This season, many families will gather and tell stories about the people who came before them. That is how legacies are carried: through retelling, and also through disagreement, correction, and the occasional “that’s not how I remember it.” The conversation is part of the legacy, not a threat to it.
AI is a tool, and a tool is neither good nor bad on its own. What matters is how a family chooses to use it, and whether that choice is made openly, with every generation at the table. Used thoughtfully, it may help families preserve stories, documents, and voices that would otherwise fade. Used without reflection, it can quietly shift who is heard and whose view carries weight. The research reminds us that preserving a voice and letting it decide are different things.
A practical place to start is an ordinary conversation. Before any such tool is considered, ask your family what they would want preserved, what they would want left to interpretation, and who should have a say. Writing the answers down, even in a single page alongside your charter or constitution, is a modest first step toward that conversation.
A founder’s legacy is strongest when the next generation is free to carry it forward in its own voice.
Research behind this issue
Hollanek, T., & Nowaczyk-Basińska, K. (2024). Griefbots, Deadbots, Postmortem Avatars: On Responsible Applications of Generative AI in the Digital Afterlife Industry. Philosophy & Technology, 37, Article 63.
Kozlovski, A., Harbinja, E., & Dobbe, R. (2026). The Many Faces of Indeterminacy in Interactive Deadbots. Philosophy & Technology, 39, Article 74.