The Best AI Companion Should Help Us Think, Not Simply Think for Us

A true story I personally witnessed. Two people were playing Monopoly, one a baby boomer and one a millennial. They needed to calculate 10% of $330.

Instead of working it out, they asked Alexa.

Alexa could answer immediately: $33. But was providing the answer the most helpful thing an intelligent assistant could have done?

That small moment raises a much larger question. As artificial intelligence becomes part of everyday life, should it eliminate every mental effort, or should it sometimes help us exercise the abilities we want to preserve?

This question is particularly important as America’s population ages.

The growing dementia challenge

A major 2025 study followed more than 15,000 adults and estimated a 42% lifetime risk of developing dementia between ages 55 and 95. Risk remained relatively low until later life, then increased substantially after age 75. The researchers also projected that annual new dementia cases in the United States could rise from approximately 514,000 in 2020 to about 1 million by 2060.

The estimate should not be interpreted to mean that 42% of everyone currently over 55 already has dementia. It is an estimated lifetime risk across the period from age 55 through age 95. Even with that qualification, the finding represents an enormous challenge for individuals, families and caregivers.

The research was published in Nature Medicine and summarized by the National Institutes of Health.

Technology cannot promise to prevent dementia. However, it can be designed to support continued conversation, reminiscence, reasoning, independence and meaningful human connection.

That is the opportunity for Voice of Reason, now Legavo.

Is AI causing cognitive atrophy?

“Cognitive atrophy” is an understandable description of what people fear, but it is not currently an established medical diagnosis caused by AI. There is not sufficient evidence that using Alexa, ChatGPT or another assistant directly causes structural brain atrophy or dementia.

There is, however, a legitimate concern about cognitive offloading.

Cognitive offloading happens when we transfer a mental task to an external tool. We have always done this. We write shopping lists, use calendars and record telephone numbers. These tools can free our minds for more valuable activities.

The problem is not offloading itself. The problem arises when technology consistently removes opportunities to recall information, estimate quantities, evaluate claims, make decisions or work through manageable problems.

A 2025 study of 319 knowledge workers found that greater confidence in generative AI was associated with less reported critical thinking. The researchers did not conclude that AI damages the brain. Instead, they found that AI changes where people apply their mental effort, often shifting it toward checking and integrating AI-generated information. The study is available through Microsoft Research.

A separate 2025 preprint examined brain activity during an AI-assisted essay-writing task. It reported lower neural connectivity and weaker recall in the group initially using an AI assistant. However, the experiment was small, focused on one specific task and had not yet established long-term effects. It should be treated as an early warning worth studying, not proof that AI causes cognitive decline. The paper is available on arXiv.

The scientific conclusion is therefore nuanced:

• AI has not been shown to cause dementia or literal brain atrophy.
• Habitual dependence may reduce how often people practice certain cognitive skills.
• AI can also support learning, independence and engagement when it is designed appropriately.
• Long-term outcomes will depend partly on whether AI replaces human thought or helps stimulate it.

The problem is not asking Alexa once

Nobody is likely to experience cognitive decline because they asked for the answer to one Monopoly calculation.

The concern is the cumulative design philosophy behind thousands of such interactions.

If technology immediately handles every calculation, remembers every detail, selects every option and completes every decision, people may gradually receive fewer opportunities to exercise everyday abilities.

This applies to boomers, millennials and younger generations alike. The stakes may be higher for older adults, but overdependence on automated answers is not an age-specific issue.

Most digital assistants are designed to minimize the time between a question and its answer. Legavo Companion can follow a different principle:

Provide the least assistance necessary for the person to succeed, while always making direct help available when it is genuinely needed.

From answer machine to cognitive companion

When someone asks, “What is 10% of $330?” Legavo does not always have to respond immediately with “$33.”

It could first say:

“Would you like a quick way to work it out?”

If the person agrees, Legavo might continue:

“Ten percent means one-tenth. To find it quickly, divide the number by ten, or move the decimal point one place to the left. What would that make $330?”

If an approximate answer is enough, it could offer an even simpler strategy:

“Ten percent of about $300 is about $30, so the exact answer should be a little more than $30.”

The user remains in control. If they say, “Just give me the answer,” Legavo should answer directly. If they appear confused, frustrated or distressed, the system should provide more assistance rather than turning the interaction into an unwanted test.

This creates an assistance ladder:

1. Offer a simple strategy.
2. Give one useful hint.
3. Allow time for the person to respond.
4. Reinforce the successful reasoning process.
5. Provide additional help or the complete answer when requested.

The objective is not to quiz, embarrass or diagnose the user. It is to preserve participation.

How Legavo can support everyday thinking

This approach can extend beyond arithmetic.

Recall through meaningful conversation

Instead of immediately supplying every personal detail, Legavo can offer gentle cues grounded in the person’s consented memories and photographs:

“This was taken near the lake. Do you remember who went there with you?”

If recall does not come, Legavo can provide the known information without judgment.

Reminiscence with photographs

Family images can become starting points for stories rather than passive slideshows. Legavo can ask about the people, places, emotions and events connected to an image, then preserve the user’s account as part of a developing life story.

Images must be treated carefully. Legavo should distinguish between verified identities, tentative suggestions and unknown people. It should never confidently invent a name or relationship.

Guided practical reasoning

For everyday, low-risk questions, Legavo can explain a manageable process before revealing the result:

• Estimating a restaurant tip
• Comparing two prices
• Working out an approximate travel time
• Following a recipe proportion
• Planning the steps needed to complete a household task

For urgent, safety-critical or medically important questions, the priority changes. Legavo should provide clear information immediately and direct the person to appropriate human assistance when necessary.

Active storytelling

When helping create a memoir chapter, Legavo can invite the person to recall sensory and emotional details:

• What did the place look like?
• Who was present?
• What happened next?
• Why did the event matter?
• Is this something personally remembered, learned from family or reconstructed later?

This exercises narrative thinking while preserving identity and family history.

Connection rather than substitution

Legavo can also help turn an AI interaction into human contact. A remembered story can be shared with a family member. A photograph can prompt a caregiver message. A memoir chapter can create a reason for relatives to call, respond and contribute their own memories.

The goal should not be to replace people. It should be to support the individual between visits and create more opportunities for meaningful connection.

Dignity must come before cognitive exercise

A companion designed for older adults cannot treat every conversation as a brain-training session.

Sometimes a person wants an answer, not a lesson. Fatigue, vision, hearing, anxiety, pain and cognitive impairment can all affect the amount of effort that is appropriate at a particular moment.

Legavo should therefore adapt to the individual:

• Ask permission before providing instruction.
• Respect requests for direct answers.
• Avoid a patronizing or corrective tone.
• Use familiar language and short steps.
• Allow extra time for responses.
• Never shame someone for forgetting.
• Reduce demands when confusion or distress is detected.
• Let users and caregivers configure the preferred level of assistance.
• Clearly separate companionship from clinical assessment.

This is especially important for people living with dementia. The purpose is not to prove what someone cannot do. It is to help them participate in what they still can do.

A better measure of intelligent assistance

The success of an AI companion should not be measured only by how rapidly it produces answers.

We should also ask:

• Did the person remain involved in solving the problem?
• Did the interaction encourage recall or reasoning?
• Was the person able to complete the task independently?
• Did the system preserve dignity?
• Did it create an opportunity for conversation or human connection?
• Did it provide direct assistance when that became necessary?
• Were its personal claims grounded in verified information?

Legavo should be evaluated against these outcomes. Its cognitive-support approach should also be tested prospectively rather than marketed as a proven method of preventing dementia.

Technology that preserves participation

The Monopoly calculation is simple:

Ten percent means one-tenth. Dividing $330 by ten gives $33.

The more important calculation concerns what we gain and lose when machines answer every question for us.

AI can become an answer vending machine that removes effort from daily life. Or it can become a patient cognitive partner that offers hints, encourages participation, supports memories and steps in fully when needed.

Voice of Reason, now Legavo, is being developed around the second possibility.

A truly intelligent companion should know the answer. But it should also understand that sometimes the most helpful response is not to give that answer immediately.

Sometimes the better response is:

“Would you like me to show you a quick way to work it out?”