Memory helps people feel understood. This can happen when a friend inquires about something you mentioned in the past. It can happen when a therapist recognizes a repeating pattern, or when someone recalls what makes you feel better after a tough day.
Though memory does not create relationships, it helps provide consistency in interactions. Without it, every conversation is a separate and inconsequential event. With it, the things people have said or done in the past are still important.
This provides a foundation in understanding the difference between AI companionship and chatbots. A chatbot can reply to requests instantly, but true companionship depends on the ability to carry personal information and context.
An example of this is Macaron. This personal AI agent seeks to remember user preferences and routines. It helps provide support in context more informed and relevant. Macaron is part of a larger study, Mind Lab, which is focusing on the development of Artificial Intelligences more interactive and responsive to real world activities, as opposed to being reliant on unique response generation.
It is this sense of being remembered and the ability to store relevant information that helps create context and continuity in a support system, that makes the difference in AI companionship.
Why Memory is Important to Emotion
The emotional presence of memories is often largely underrated. For someone to hold onto a memory means that something about us was good, worthy and valuable enough to keep. That something about our being held value did not just disappear when we left or when we stopped talking.
Recognition is important to human relationships. We don’t feel recognized simply because someone has memorized everything about us. Rather, we feel recognized when someone has remembered the details that matter most to us. This could be remembering a related tough topic we discussed, or recalling that we were nervous about an event that was coming up.
Unlike humans, AI does not remember, it does not feel like we do, and it does not form emotional attachments. This does not mean that memory does not have the potential to change user experience. Interaction does not have to be generic. Memory does have the ability to make user experiences feel specifically tailored to the user’s actual lived experience.
This could all help make AI even more helpful and comforting for users because many people use AI during every day moments of planning, organizing, reflecting and trying to make sense of the week that has been stressful. Memory can help contextualize user support so it does not feel like random or generic support.
Memory is More Than Saving Chat History
It is crucial to differentiate between the concepts of remembering and saving. A system could save a long history of conversations and still fail to utilize that information. Remembering is not about retaining everything. Remembering is about providing the right context at the right moment.
This is where the technology of personal AIs becomes applicable. An AI companion, to be perceived as helpful in a long term horizon, needs more than a congenial approach. It needs the ability to connect past information to present situations. An AI, if the user has expressed that morning is a time of great stress, and if the user prefers a short and efficient morning routine, and if the user has stated that he/she wants to improve their routine and/or meal schedule, then that AI should take all that information into account when it comes time to offer a suggestion.
If, for example, the user says, “I feel scattered tonight,” a generic AI would respond with broad and likely meaningless suggestions about how to relax. An AI that possesses a more refined sense of memory would probably recall that the user has been trying to build a more positive evening routine, and the AI would suggest the user take the next step in that routine by preparing breakfast for the next day, or perhaps taking a moderate evening stroll.
This is a very practical suggestion that improves the quality of the memory-based relationship. It is not therapy. It is not drama. It is not emotional intelligence in the sense of a human being. It is more connected to the individual’s life and therefore, more useful.
It is particularly applicable in this case because human beings are creatures of memory. Human beings do not experience life in a collection of isolated events that have no relationship to one another. Humans are memory based creatures and therefore, for a long-term relationship, an AI has to be ‘memory based’ and ‘relationship’ based.
People may be able to identify patterns in their behaviors and emotions that they didn’t notice before. For example, you may notice that you feel anxious as the weekend comes to a close. You may even be able to identify that your work schedule impacts your eating habits. Maybe you only eat lunch when you have time to. Perhaps you have a pattern of not eating at all when you have a busy work day. You may even feel better when you write out your schedule before going to sleep. It is harder to identify patterns when you are busy and life is chaotic.
A memory conscious AI can help organize the patterns. It may say, “I’ve noticed you feel overwhelmed before the new week begins.” It may say, “You feel your best when you plan dinner ahead of time.” These may help the user to be reflective.
The most important aspect is the AI needs to be careful about what it says. It shouldn’t say, for example, “I have a full understanding of your emotional life,” or other over claims. It may help you recognize patterns you haven’t noticed before.
Memory aids and encourages self awareness. It helps you by providing a reflection, not a diagnosis.
Why Saying The Same Thing Over And Over Can Be So Exhausting
Recounting the same thing over and over again is emotional and physically draining. This is most true when it’s something personal. Repetitively saying what you like, what you prefer, or what you simply want, or even your emotions, can be very mentally draining.
This is when memory becomes most useful. Memory integrates useful contexts and the user doesn’t have to go back and explain the whole thing and can keep talking where they last left off.
Continuity is beneficial in many day to day situations. For example, someone trying to improve their habits doesn’t want to explain each time s/he is not a fan of strict meal plans. A student probably doesn’t want to tell each person he/she comes into contact with, that he/she studies best through short bursts. In the same vein, a parent probably doesn’t want to update the family schedule every single week.
Although these scenarios are practical, they are also emotional. For instance, the absence of the need to modernize one’s explanation, makes the interaction less formal. It feels less like the interaction is with a tool, and more like a memory support system.
The Next Step: Real Life Learning, Not Just Response Learning
The beauty of personal Ai systems is that they don’t just have to get better with answering questions, but that they can actually learn about the way a person lives and the way that Ai companionship should be, the way that life is: chaotic, and ever changing.
Makarons’ philosophy is interesting. It is focused on the conversation and communication aspect of Ai systems, yes, but it also incorporates the day-to-day tools and routines, and on the research side, systems of Ai that can learn and adapt through real life feedback and real life engagement.
For users, a product with this level of sophistication will build a general understanding of what types of reminders and tools are actually helpful, what routines and tools are feasible to integrate, and what incorporation and engagement styles will actually lead to sustained usage. Systems that learn from behavior create more personalized and less generic experiences.
Learning from users needs to be done with care. It should be done with the utmost concern for privacy, protection, and control of the user. A boundary-less remembering companion will become quickly and justifiably seen as an invasive tool. The purpose of a good remembering companion is to support, and not define, control, or limit.
AI Companionship Is Still Not Human Companionship
This needs to be stressed and kept in mind when interacting with systems that remember. Being remembered by a system stands in stark contrast to being known, loved, or cared for by a fellow human being. Relationships are a two-way street: both have the capacity to be affected, and have the mutual willingness to be vulnerable to one another. AI has no capability to act in this manner.
AI systems that remember your habits and respond in a friendly manner are still tools. Healthy AI companionship should not be a substitute for interpersonal relationships, and should encourage the user to pursue meaningful relationships outside of the community. AI companionship should promote planning and help organize the day. It should leave space for community, therapeutic, and family support. This is particularly true when someone is dealing with serious mental health concerns, trauma, or emotional distress.
AI companionship is best used in a supportive role and should help users identify patterns, structure their day, and help them refocus on their goals. AI can help streamline day-to-day activities, but it cannot fulfill the type of companionship and mutual care that only people can provide.
The Dangers of Excessive Memory
Memory is certainly a gift, but it is also a burden. Not every detail needs to be kept. Not every pattern needs to be reinforced. Users should be able to see and change what is remembered, and be able to control the memory and delete it when necessary.
There is also a threat of narrowing the focus in memory via excessive personalizations. If AI only adapts to what someone has done, it can trap them in a cycle of old habits. A good memory should not create a prison.
Take for instance someone who has a hard time engaging in what they see as tough conversations. This should not be avoided. It is more important for the memory system to help the user bring out their goals of having of having difficult conversations or helping them engage in tough conversations more effectively, as a way of helping them become a better more confident communicator.
Memory, when done right, helps promote progress and growth, as opposed to stagnation.
The Importance of Memory
Memory is important to establish a sense of connection and help not lose sight of the things that other people hold valuable in the present. In AI, this can help make what would seem like routine supports relevant and grounded.
The true value of memory is not in imagining AI as a human; it is in the ability to create technology that is more responsive to the real world. A memory-aware AI can, in a responsible design, help a user identify and understand patterns, and assist in establishing and maintaining routines so they feel a little less like they have to start from square one each time (which is quite the advantage).
What memory-embedded AI companionship can offer us is not the ability to substitute for human companionship (which would be a terrible thing). Rather, it provides the opportunity to help us offer ongoing, individualised, and patterned support throughout the day.
Leave feedback about this