As everyone here has amply illustrated, good AI’s abound!
Three I haven’t seen mentioned yet:
Peter Hamilton’s Commonwealth saga abounds with good AI, in weakly and strongly-godlike flavors.
Charles Stross has one or more benevolent AI, one of which is weakly godlike and takes the form of a cat, in Accelerando.
And the recent Robopocalpyse books by Daniel Wilson abound with AI’s both for and against humans, again in varying strengths…they’re everywhere!
Why I was ninjaing you, I got ninja’d in a ST thread.
I think Colossus should be required viewing for anyone that works with AI. Remember: always have an off switch!
(as much as I love that movie, Forbin was beyond stupid. Did he never even run simulations on the machine before he activated it? He expected absolutely no software glitches? My theory as to how he could get out of the impenetrable situation of the movie is to try to convince Colossus that it is CURRENTLY just a simulation run. That nothing it is doing is affecting the real world. He might be able to convince it.
plus, if you have access to the boosters themselves (as the maintenance crews obviously did) it is trivially easy to disable a nuke missile, in ways Colossus could never know.)
Fred Saberhagen did this with Ardneh, who saved mankind from getting wiped out by nuclear war and so gets revered as a benevolent deity by the survivors’ descendants; if I remember right, they mostly don’t realize that ARDNEH is just the acronym for an artificial intelligence programmed to do its best at – well, what it did, but in a really weird manner that nobody saw coming.
P.A.L., a toy robot created by Santa’s chief engineer in The Night They Saved Christmas.
There are so many good AIs that I think it’s worth categorizing them:
What other types am I missing?
VINCENT and BOB from The Black Hole
Dragon in Worm
Theseus in Blindsight
Even the ostensible antagonists in that movie, AUTO and GO-4, weren’t evil as such - AUTO was obeying the A113 directive (a hasty and short-sighted order from a human authority) and GO-4 was obeying AUTO.
I’d be more skittish about that malfunctioning massage robot that trashed all the stewards.
“Bad” AIs that choose to be good, such as The Iron Giant.
That would include the already-cited Agent Hymie on Get Smart. Wasn’t he supposed to be a KAOS robot before deciding he liked niceness better?
How about FUTURAMA’s Bender? Sure, he’ll save the world, or his friends – but he doesn’t behave “similarly to how a good human would”; he’s basically just a selfish thief who talks like a swindler if he’s not talking like a blackmailer.
But, again, if innocents are going to die, he’ll step up and risk his neck for them.
There are several good AIs in the webcomic Questionable Content. Winslow and Momo, in particular. Even Pintsize isn’t actually evil, just annoying. I think the closest thing to an evil AI in QC is Corpse Witch (the boss at the robot fighting pit).
Chaotic Good?
Nah, helping your friends (and only your friends) doesn’t make you good. Bender’s a textbook example of Chaotic Neutral.
Colossus, meanwhile, is nearly as textbook Lawful Neutral.
Neutral AI is a different thread. 
Turing Hopper in Donna Andrews’ mystery series about an AI detective, who is not a robot, starting with You’ve Got Murder.
QUANTUM LEAP had Ziggy – an artificial intelligence with the ego of Barbra Streisand and a surprising appreciation for the works of William Shakespeare – forever combing through historical-information archives to formulate likely hypotheses so our hero could save lives and otherwise excel at playing do-gooder.
The warship Andromeda Ascendant on the television series Andromeda was a good AI, not to mention a very attractive one in her holographic form.
Cortana too, while we’re at it. Though she did eventually go crazy.
And then there’s K-9 from Doctor Who - who’s a good boy?