Future of AIAI & Technology

Intelligence Shaped by the Life You Live: The Future of Personal AI

By Yang Sun, Co-founder and CEO of Looki

Artificial intelligence can summarize complex documents, analyze information and solve increasingly sophisticated problems. Yet despite these advances, the way most people interact with AI hasn’t fundamentally changed.

We open an application, type a prompt, explain what we need and wait for a response. If something is missing, we provide more context and try again. For all of AI’s capabilities, we’re still doing much of the work required to make it useful.

What if the next major breakthrough in AI isn’t about getting better answers, but about needing to ask fewer questions?

During my time working on Google Assistant, the technology operated very differently from today’s generative AI. Traditional assistants relied on structured systems that identified a user’s intent and routed it to a specific function. Each new category of tasks required dedicated development, limiting what the technology could accomplish beyond its predefined capabilities. This helps explain why smart speakers largely settled into familiar uses such as playing music, setting timers and controlling smart home devices.

Generative AI has expanded those capabilities dramatically, but the interaction model remains largely reactive. People still act as translators between their lives and AI, converting what they see, hear and experience into instructions a system can understand. I believe the next evolution is AI that can perceive more of the world alongside us, recognize what matters and identify opportunities to help without always waiting for a prompt.

Intelligence Without Context Has Its Limits

AI systems can process vast amounts of information, but there’s a fundamental gap between what they know about the world and what they understand about the people using them.

An assistant might offer thoughtful advice about a difficult workplace conversation without knowing the relationships, previous interactions or circumstances involved. Its recommendation could be technically sound while missing details that fundamentally change the situation.

Much of human experience happens outside digital interfaces, through conversations, changing environments and everyday activities that AI systems currently have little visibility into. Bridging that gap requires a richer understanding of real-world context.

I believe three elements will be particularly important: visual context, audio context and continuity over time. Visual information helps AI understand physical surroundings, while audio adds conversations, tone and environmental sounds. Together, they provide a shared frame of reference that language alone often cannot convey.

Continuity connects those observations into something more meaningful. Seeing a prototype, hearing a discussion about a problem and later observing a revised version provides far richer context than any single moment. Instead of treating each observation independently, AI needs to understand how experiences relate and develop over time.

But capturing more information isn’t necessarily the answer. Continuous first-person video, for example, contains long stretches of repetitive information interrupted by brief but meaningful changes. Processing every observation with equal attention would be inefficient, while discarding information too aggressively risks losing details that become important later.

The challenge is determining what to preserve before knowing exactly how that information might be needed. What matters varies between people and changes over time. Building useful personal context therefore requires systems that can compress redundant information, preserve meaningful details and retrieve them when relevant.

From Reactive Tools to Proactive Assistance

Digital assistants have traditionally followed a familiar model: the user asks and the technology responds. More context-aware AI could change that relationship by recognizing situations where assistance might be useful.

I experienced a glimpse of this after landing at Shanghai Pudong Airport. As I approached a ticket machine to purchase a maglev train ticket, an AI assistant reminded me that I could receive a discount using my boarding pass from that day.

The system connected two pieces of information: I had recently arrived on a flight, and I was now approaching the ticket machine. Neither observation was particularly useful on its own. Together, they created an opportunity to help at exactly the right moment.

I hadn’t opened an application or asked how to save money. Any capable AI model could answer a question about the discount. What made the experience different was recognizing when that information mattered without requiring me to ask.

This is the potential of proactive assistance. An AI system might help someone recall a detail from an earlier conversation, recognize a recurring commitment or connect information across separate experiences. Its value comes not simply from having the right information, but from understanding when that information becomes useful.

However, proactive assistance also requires knowing when not to intervene. Someone entering a restaurant might be there to eat, collect an order or meet a friend. Acting on an incomplete understanding could quickly turn a helpful suggestion into an unwanted interruption.

The distinction comes down to intent, timing and control. Offering relevant information leaves the decision with the user. Purchasing something, contacting another person or making a commitment requires a different level of authorization. Sometimes the most intelligent response is to stay quiet.

What Needs to Change Behind the Technology?

Moving beyond prompts will require more than increasingly powerful language models. It depends on several capabilities working together: perception, memory, contextual reasoning and judgment.

The first challenge is enabling AI to understand the physical world. New hardware and interfaces can provide access to visual and audio information, but devices must also be practical for everyday use. Wearable technology, for example, needs sufficient battery efficiency and comfort to remain available throughout the day.

That information must then be interpreted. A system might recognize that a conversation occurred without understanding its significance or detect a change in visual surroundings without knowing whether anything meaningful happened.

Continuous visual and audio streams present a particularly difficult problem because important events aren’t always obvious. Turning your head can completely change the camera image while the underlying situation remains the same. Meanwhile, an important conversation might begin with almost no visual change.

Memory introduces another challenge. AI must preserve enough information to support future needs without retaining every observation indefinitely. Unlike text, where someone has already selected information to communicate, continuous sensory data requires the system to identify relevant details from large amounts of repetitive input.

This makes efficient compression and retrieval essential. A useful memory system must preserve factual and temporal details while allowing AI to identify which parts of the past matter in a new situation. It must also distinguish between observed facts, inferred meanings and uncertain conclusions.

Consider a discussion about a proposal that has been postponed until new data becomes available. If the topic comes up again, an assistant could retrieve that earlier condition and provide relevant context without requiring everyone to reconstruct the previous conversation.

But understanding context is only part of the challenge. AI also needs judgment about whether assistance is wanted, whether its interpretation is reliable, and whether the timing is appropriate. Sometimes it should offer a suggestion, sometimes act within permissions the user has already granted, and sometimes do nothing.

I believe the industry is underestimating how difficult it is to make that entire process work reliably. Progress won’t be measured solely by the amount of information AI can process, but by how effectively it transforms everyday experiences into useful, retrievable context.

Less Human Input Shouldn’t Mean Less Human Control

As AI becomes better at anticipating needs, an important question emerges: How can these systems become more helpful without reducing user control?

For me, meaningful control begins with a simple choice: when is AI present, and when is it not? People should be able to determine when a system can see or listen, clearly understand when those capabilities are active and turn them off easily.

The same principle applies to memory. Users should be able to review what AI remembers, correct inaccurate information and delete details they no longer want retained. An assistant’s understanding of someone should evolve with their circumstances rather than becoming a permanent record of past assumptions.

Automatic should never mean unauthorized. An assistant might recognize an upcoming appointment, but that doesn’t mean it should automatically have permission to contact someone or rearrange a schedule. Understanding a situation and having authorization to act are separate things.

AI systems will also make mistakes. They may misunderstand conversations, overlook relevant details, or draw conclusions that aren’t supported by the available information. When that happens, the system should acknowledge the error and allow the user to correct it.

Importantly, corrections should carry forward. If an assistant misunderstands someone’s circumstances, updating a single response isn’t enough. The underlying memory or assumption should also be corrected so the same misunderstanding doesn’t continue influencing future interactions.

There will also be situations where information has been lost or was never captured. Rather than inventing an answer, AI should acknowledge those limitations. Trust depends not only on getting things right, but on responding reliably when something goes wrong.

The goal is to reduce the effort required to use AI without reducing the user’s authority over it.

The Best AI May Be the AI We Have to Manage Less

Prompts aren’t going to disappear, nor should they. Explicit instructions will remain valuable for complex decisions, creative work and situations where people want direct control. But they shouldn’t be the only way to benefit from AI.

Over the next five years, I expect more interactions to begin with AI recognizing an opportunity to help. It might recall something relevant, connect separate experiences or prepare information before someone explicitly requests it.

That understanding could also extend across physical and digital environments. Something discussed over lunch might become relevant when someone returns to their computer, with different devices and AI agents contributing to a more continuous understanding of their activities.

The biggest challenge may be less about building increasingly capable agents and more about ensuring they receive the right context at the right moment. A system that understands what happened, what changed and what matters to someone can offer assistance that feels more natural and less demanding.

For years, AI progress has been measured by what systems can accomplish when given the right instructions. The next chapter may require asking a different question: How much effort should people have to invest before that intelligence becomes useful?

AI’s future may not be defined by how much we can ask it to do, but by how much it can help us accomplish without having to ask.

Related Articles

Back to top button