Jason Calacanis, serial entrepreneur, just launched Mahalo, a search engine where users get hand-crafted portal-like results for common search queries. It's based on the theory that many people are searching for the same things, that search engine spam is making Google less useful for common queries, and that humans are still wiser than algorithms at sifting through results and finding the really good stuff. Essentially, the site plans to have employees (along with a dash of user-submitted content) build a portal of links to the best information for each popular search term. If a portal doesn't yet exist for a search term, Mahalo returns Google results by default.
The idea caught my eye as innovative. It might even work. So here's my feedback to Jason: it seems to me that where Mahalo really has the potential to shine (yet is only doing a middling job at, currently) is in giving context. Half of the human-powered-search equation, and most of the work, is having real people do the research and come up with the best links on a topic. The second half is giving people enough context to know what they should click on. Mahalo has lists of great links, but it also needs to tell the story of each link. It'd only take 3-5 words each, max. A link without a story about what you'll find there and why you should take the page seriously isn't any better than a random Google result.
A lot of entries do pretty well at this- air conditioner, sake, cholesterol management; some don't- yoga, obesity (I'm mostly going off the Top 7 links). But since giving intelligent link context is something human-powered search could do really well and is really helpful for people, and also something that Google can't do, I think it'd be worthwhile for Mahalo to make it a top priority.
So, that's the angle I'd suggest for competing against Google. With a short (3-5 word) story about each link, search results for common queries could be better, more contextual, and more concise. Competing against Wikipedia-as-a-search-engine will actually be tougher, I think, since Mahalo's portal-style search results and Wikipedia's summary-style articles occupy overlapping niches.
Finally, some search engine diversity!