What Does The Ending Of Ex Machina Really Mean?

What Happened To The Other A.I.?

Just as Ava wasn't going to be the last A.I., neither was she the first. At the end of the film's second act Caleb discover footage of Nathan's earlier attempts, which failed in rather violent manners.

The film doesn't dwell on what exactly went wrong, but as they were concerned with escaping the problem doesn't seem to be lack of awareness. Rather, it was rationality - whereas as previous A.I. thought the only way to escape was by hammering at the door incessantly, Ava understands she needs to think to get out.

One area where Ava is seemingly less advanced is her aesthetic design - she's clearly a machine, with no artificial skin - which actually serves a practical purpose. Not seeing any of the underlying mechanics makes it easy to forget the android underneath the skin, meaning the passing of the Turing test comes more from look than the A.I. itself. As Nathan stated, the constant reminder was needed.

Who Was The Assistant?

The only other A.I. seen alive during the film's narrative is Kyoko, Nathan's assistant, revealed to be an automaton in a gruesomely absorbing skin peeling scene. How complex she is isn't made clear, although it's likely she has some form of inhibition - otherwise she would be allowed to roam the house freely. Originally introduced as a human employed by Nathan to look after his house, the fact she's a robot serves as a further view into Nathan's character.

There's a heavily implied sexual relationship between the two, which plunders Nathan's subconscious loneliness, but also points to his conflicted relationship with his creation - he regards Kyoko with consistent distain, treating her as decidedly inhuman. Their final interaction - Kyoko stabs Nathan - could suggest that she was in fact capable of making decisions and thus of a higher intelligence, but she's really only a blunt instrument, doing it under Ava's suggestion.

Despite not being able to speak (to maintain the illusion of being a human without a grasp of English), following Ava's orders means she can comprehend it. This highlights a rather big flaw in using human avatars for artificial intelligence - we measure them only by their ability to mimic us and their power of speech, rather than in more varied terms (as the original Turing test posited).