Google simply launched its most spectacular (and unsettling) addition to AI Overview but, a brand new characteristic referred to as “Audio Overview” that generates audio summaries of search outcomes, narrated within the type of two life-like, but not fairly human podcast hosts.
Audio Overview is presently an opt-in Search Labs characteristic, which means you will not see the choice for it until you toggle a change in Search Labs. Proper now it is solely accessible within the U.S. and solely generates English summaries.
I attempted out Audio Overview myself and the outcomes weren’t precisely what I used to be anticipating.
After you activate the characteristic in Search Labs, some Google searches will embody an Audio Overview field, normally beneath the common AI Overview and “Individuals additionally ask” sections. You simply faucet the button to generate the audio abstract and look forward to it to complete processing.
The audio clip you will get is generated on the spot, so for those who refresh the web page and generate it once more, it may find yourself being completely different. The summaries I generated ranged from 3 to five minutes lengthy. All of them characteristic the identical pair of AI-generated voices who travel discussing no matter matter you searched, within the type of a podcast.
The voices are admittedly considerably extra lifelike than the robotic Siri sound I used to be anticipating. There’s tone adjustments, conversational phrase selections, seemingly pure language. It isn’t fairly practical, although. The 2 voices are like a pair of podcasters with zero rapport who appear to be they’re studying off a teleprompter. It is simply shy of being lifelike however nonetheless lifelike sufficient that some folks might be fooled at first.
Google exhibits you which of them search outcomes it used to generate the audio abstract, so you possibly can double-check no matter information your AI podcasters provide you with. Nonetheless, they sound practical sufficient that some folks may simply assume these are actual folks and take no matter they are saying as truth. In fact, that is additionally a problem with textual content AI summaries.
There are some hiccups that give away that these aren’t actual folks. As an example, in one of many summaries I acquired, one of many AI voices asks a query then instantly solutions it herself, which sounded fairly awkward.
Each voices use emotional language sometimes, like exclaiming “Wow!” at a enjoyable truth, but it surely undoubtedly sounds stiff and simply shy of one thing an actual individual would say. The AI voices additionally mispronounce phrases every now and then, like “musk” as an alternative of “should.”
Uncanny and eerie as this characteristic is, I can see it being useful to some folks, particularly those that could have imaginative and prescient impairments, or in any other case depend on display studying instruments. The AI-generated voices sound fairly good for the truth that they’re AI, too.
That will be cool if it did not include a bunch of considerations across the unfold of misinformation by way of AI and the risk AI-generated voices like this might pose to jobs like voice appearing. Like all innovation in AI, Google’s Audio Overviews are a double-edged sword and sadly I am nonetheless extra skeptical than impressed.
