Memento (Film)

By ai-depot | June 30, 2002

We’re proud to announce our first movie review here at the AI Depot. And what a way to start? A mind-bogglingly clever, entertaining film that has numerous deep insights into human intelligence and the importance of memory.

Written by Alex J. Champandard.

Film Review

Starring: Guy Pearce, Joe Pantoliano, Carrie-Anne Moss
Director: Christopher Nolan
Year: 2000
DVD: Available at Amazon US and UK.
VHS: Available at Amazon US and UK.

Overview

Memento is a stunning movie, amazing acting, cleverly directed, refreshingly different, and surprisingly funny. Not only will it keep you entertained for a couple hours, but you’ll be talking about it for days after, and pondering about it for weeks…

This is one of my top 5 films ever, and even though there are about 10 in that category, this one actually is in the better part of the top 5 ever!

Memento Film

As we will discuss, the film has deep insights into the current state of the art of artificial intelligence. This may come as a surprise to you, but this is in fact part of my inspiration for my post-graduate research!

The Premise

I’m not going to reveal the plot here! So don’t worry about reading this section… I’ll only portray the settings of the film.

This is the story of a man whose wife gets killed, an event which has deprived our hero from the ability to make new memories. Having worked for an insurance company, the man has experience in the area: a previous client of his has a similar disability. He manages to keep himself on track by passing information to himself, using an organised system of tattoos and Polaroids. And that’s pretty much where the film starts.

Skip to the artificial intelligence section now if you don’t want to know anymore about the film itself!

The Structure

This is truly where the film shines, based on an inspired editing technique. Getting the impression of loss of memory across is very difficult, but the director manages to achieve this by moving backwards through time! This mostly is done in ten minutes chunks, so the story itself spans a couple of days. It’s a bit like having a plot twist every ten minutes! Very entertaining. Admittedly its quite confusing the first time it happens, but the director manages to pull this off really well using a fade to black and a couple seconds overlap in the story. You’ll get use to this if you pay attention and are open-minded enough to accept this novel way of story-telling.

Those people that disliked the movie’s structure claim this was the director’s attempt to put his name to the first ‘backwards’ film. This is completely ridiculous as it’s already been done before, and the goal of the film was to get across the loss of memory factor.

The Story

Once again, I recommend watching the film first before you read this section, so skip along to the artificial intelligence part of the review if you haven’t!

This is the amazing part folks: the story goes backwards, and still the plot thickens! But rather than watching it unfold — like your average film, you find out how it was wrapped up in the first place. This concept provides the director with so many new opportunities, which he takes advantage of in a very skilful way… as we discover the past of the main character, the scenes we have already seen take scope and become seen under a new light. In particular, distinguishing which people are simply taking advantage of this man, and which people are actually trying to help him!

The film does finish very well if you follow the plot. You knew the ending right from the start, but you find out who was who, and what they did in the past that caused the events that happened in follow-up. All in all, the only one detail is left unsure: what exactly happened to his wife. The rest becomes crystal clear in the last scene.

The way the story is done also provides good re-viewing. You’ll pick up more and more as you watch the film again and again. And that’s another amazing aspect of the film.

Artificial Intelligence

So what makes this film belong on an artificial intelligence site? Well there are three major things to think about, all of which link together in a way that is ideal for implementing deliberative intelligence within a smart agent.

Reactive Behaviour

Your average animal has no ability to take into account the environment in a complex fashion. It is limited to simple behaviours that are triggered by events and sensations. This is known as purely reactive behaviour, as the animal simply reacts to its environment without any further reasoning. This can be tested by submitting the animal to the same conditions over and over, and analysing its behaviour. Some animals can learn simple behaviours, which can cause them to react in a different fashion, but this has very little impact on their ability to concoct elaborate plans.

The main character of this story, without the help of his notes, would have his capabilities reduced in a similar fashion. He would dispose of 5 short minutes to perform any task, after which he needs to come to terms with his situation again. This fundamentally restricts his ability to do anything complex.

With the help of his reminders and tattoos, he can start to aspire to a little more than just ‘reactive tasks’.

Deliberative Planning

The ability to make long term plans is a characteristic of human intelligence. This is only possible when the person in question is capable of elaborating such a plan, and commits himself to it. This implies that future decisions and actions will be compatible with this commitment. Micheal Bratman was the first person to characterise this property of plans and intentions.

In the film, the main character gives himself the ability to do this using notes and tattoos. These bodily decorations give him the ability to set himself long-term goals, which he can attempt to carry out in 5 minute chunks. In general, the tattoos give him a starting point, a reference that will allow him to start his plans.

Emergent Intelligence

Despite him having guidelines for his plans, they tend to unfold without him being able to do much about it… you could call this emergent intelligence as he only passively contributes towards the overall goal, by unknowingly performing simple reactive tasks.

This leads into one key concept, which some researchers have started to study in various forms. With regards to integrating deliberative plans with reactive goals, one can express the high-level plans as a sequence of low-level actions. This allows an AI agent to simply take into account this reactive state, which is updated and rigged by a higher-level component in order to provide the overall emergent intelligence required.

I personally am using this in the context of robot navigation, by expressing long paths as a sequence of short relative moves. See my tutorial on robot navigation for more information.

Importance of Memory

This underlines the importance of this memory. In the case of the average humans, new memories are fabricated on demand, so the person is constantly up-to date with what his goal is, and how much of it he has achieved. For the main character in Memento, this is only done by tattoos and notes. This is incredibly error prone, and as you’ll find out towards the end of the film, voluntarily falsifying this state has important consequences. It also allows him to make what he wants of his life… he may have voluntarily lost parts of the reports to have the pleasure of uncovering them, or for example telling himself that someone else actually killed his wife.

For computers, however, an approach with such a separate memory seems ideal. As we have seen, this provides intelligent behaviour in an emergent fashion. Additionally, for a machine, the memory is as easy to deal with as the environment. There is an intense debate going on currently in AI, between those that think that agents should use the entire environment as their input (providing memory implicitly), and those that believe in explicitly remembering things within the agent’s framework.

I believe this memory to be essential, and the only link between otherwise independent moments in time… a concept which the film portrays beautifully.

Belief-Desire-Intention

All this debating leads into a new kind of belief-desire-intention framework, as originally described by Bratman. The whole system is based on a separate state, which provides the link between the deliberative thoughts, the beliefs updated by the agent based on the previous state, and the desires corresponding to the current state of the agent.

Though I’ve not admitted this to myself yet, this is exactly what my research is leading into, and no doubt you’ll be hearing more about this!

In Summary

This is a film you need to see at least twice, which probably means you’re better off buying it! This is one of my favourite films, and if you’ve read this review all the way to the end, no doubt you’re interested enough for it to be one of your favourites too!

Why not go and buy this classic over at Amazon.com now?