Google Launches AI Platform That Looks Remarkably Like A Raspberry Pi

Google has promised us new hardware products for machine learning at the edge, and now it’s finally out. The thing you’re going to take away from this is that Google built a Raspberry Pi with machine learning. This is Google’s Coral, with an Edge TPU platform, a custom-made ASIC that is designed to run machine learning algorithms ‘at the edge’. Here is the link to the board that looks like a Raspberry Pi.

This new hardware was launched ahead of the TensorFlow Dev Summit, revolving around machine learning and ‘AI’ in embedded applications, specifically power- and computationally-limited environments. This is ‘the edge’ in marketing speak, and already we’ve seen a few products designed from the ground up to run ML algorithms and inference in embedded applications. There are RISC-V microcontrollers with machine learning accelerators available now, and Nvidia has been working on this for years. Now Google is throwing their hat into the ring with a custom-designed ASIC that accelerates TensorFlow. It just so happens that the board looks like a Raspberry Pi.

On board the Coral dev board is an NXP i.MX 8M SOC with a quad-core Cortex-A53 and a Cortex-M4F. The GPU is listed as ‘Integrated GC7000 Lite Graphics’. RAM is 1 GB of LPDDR4, Flash is provided with 8GB of eMMC, and WiFi and Bluetooth 4.1 are included. Connectivity is provided through USB, with Type-C OTG, a Type-C power connection, a Type-A 3.0 host, and a micro-B serial console. Gigabit Ethernet, a 3.5mm audio jack, a microphone, full-size HDMI, 4-lane MIPI-DSI, and 4-lane MIPI-CSI2 camera support. The GPIO pins are exactly — and I mean exactly — like the Raspberry Pi GPIO pins. The GPIO pins provide the same signals in the same places, although due to the different SOCs, you will need to change a line or two of code defining the pin numbers.

You might be asking why Google would build a Raspberry Pi clone. That answer comes in the form of a machine learning accelerator chip implanted on the board. Machine learning and AI chips were popular in the 80s and everything old is new again, I guess. The Google Edge TPU coprocessor has support for TensorFlow Lite, or ‘machine learning at the edge’. The point of TensorFlow Lite isn’t to train a system, but to run an existing model. It’ll do facial recognition.

The Coral dev board is available for $149.00, and you can order it on Mouser. As of this writing, there are 1320 units on order at Mouser, with a delivery date of March 6th (search for Mouser part number 212-193575000077).

There’s also another device in the hardware portfolio, called a USB accelerator, which they can only assume is the Edge TPU connected to a USB cable. This USB accelerator will work with the Raspberry Pi — that’s from Google’s product copy, by the way — and will get you started on machine learning inferencing with the Edge TPU designed by Google. The price for this USB accelerator is $75 USD.

We would like to congratulate the Raspberry Pi foundation for creating something so ubiquitous even Google feels the need to ride the coat tails.

Fraudsters are already tricking Google’s Performance Max - what if it got worse?

Mathew Ratty, co-founder and chief executive officer at TrafficGuard, has been testing Google’s next-gen buying platform - he’s worried fraudulent signals will corrupt the machine learning system.

When Google announced Performance Max (PMax), it seemed like the answer to every marketer’s dream – drive marketing efficiency, performance and better ROI across all of Google's channels, including YouTube, search, shopping and discovery.

At the core of the PMax promise was Google’s AI, making decisions on everything from bidding to creative to search query matching and media environments. For many marketers, this even helped open the door to experimenting with innovation in online video, such as YouTube Shorts. However, like all AI and machine learning-driven advertising platforms, whether originating from Google, the Trade Desk, Yahoo, or numerous others, marketers must place trust in the enigmatic algorithmic ‘black box.’

This calls for a dose of prudent skepticism.

The problem with all black box systems is that marketers are at the algorithm’s mercy. In the case of PMax, where Google manages everything, it requires an even higher level of faith in the system. Limited granular reporting in PMax means that while you get broad campaign insights, you won’t be able to see if it’s display, search, video, or shopping ads driving your clicks and conversions. This would be great if the world was perfect or you only spent money on Google channels – but marketing is complex, with multiple media partners in any campaign. They all know from experience that industry opacity can be exploited to the detriment of marketers and their budgets.

Transparency and accountability are indispensable for marketers to operate with utmost effectiveness. They have been analyzing PMax campaigns with select clients over the past several months, and they believe the industry should know the truth about PMax and all AI-driven optimization platforms.

It should be no surprise that bad actors are shifting their focus from general programmatic fraud to targeting campaigns with no insights. The recent supply chain study conducted by ANA demonstrates waste and inefficiencies in the existing programmatic supply chain. PMax is no exception, with the same underlying risk from invalid traffic (IVT) as they have seen on traditional programmatic campaigns. For this piece, the definition of invalid traffic is traffic that neither contributes to incremental growth nor originates from human sources, encompassing entities like bots and data centers.

In traditional programmatic, invalid traffic and click fraud occurs daily, spanning search, mobile, and affiliate campaigns. For context, in search, they have seen between 5-15% of search clicks come from bots seeking to exploit paid search campaigns to sign up or claim incentives within the ads or deliberately exhausting clients’ search budgets as a “competitive” tactic. In a parallel context, not all instances of invalid traffic bear malicious intent. They have also concluded that 97% of a user’s Google ad budget was consumed by returning users using Google as a front door to click on a paid ad to log in to their account.

In the mobile domain, the figures they observe are even more disconcerting, particularly concerning for app install campaigns within sectors like car sharing and food delivery. Instances of fraudulent app installs have surged to alarming rates, reaching up to 50% and, for one client, the claimed clicks and installs exceeded the population of the targeted geography in a week.

When addressing affiliate fraud, specifically for high-payout categories such as sports and sports fantasy betting as well as subscription and entertainment services, they have seen click fraud and affiliate cookie stuffing through malevolent browser extensions siphon away $100,000 of affiliate payouts per month – harming marketers and publishers. Marketers and publishers being targeted should be concerned, especially with so many publishers pivoting to affiliate sources to offset general advertising challenges.

What they have observed on PMax is a mix of new fraud tactics and some of the same types of invalid traffic and fraud that they see across programmatic. Given that PMax, and most AI systems, assume positive user intent, invalid traffic in PMax campaigns comes because AI thinks every “user” engagement is positive in intent. When bad actors exploit this and create fake intent signals, it can train the algorithm to optimize toward the source of the invalid traffic. This results in wrongly optimized campaigns that divert and deplete advertising budgets by driving more fake engagement and conversion events.

This is not malfeasance by Google or PMax; however, when AI optimizes towards invalid traffic, budgets can be exploited when there is opacity and no real-time 3rd party oversight and intervention. PMax is simply a microcosm of what happens across the entire internet – the challenge lies in marketers who employ AI buying systems without independent third-party auditing and analysis. They face the possibility of unknowingly running the risk that they will just be pumping money into a high-speed AI-optimized invalid traffic machine.

Our trials of their PMax analysis platform focused on providing transparency of what channel type (search, shopping, display, video) drives the conversion results. While Google’s AdManager interface for Performance Max fails to display this information, they found a way to provide visibility into these insights – because transparency matters when fighting fraud. This includes an ability to clearly, by channel, for every PMax campaign, in real-time, identify where your money is going and what is driving your performance and solve for what the ANA recently highlighted as the inefficiencies and waste that would otherwise come from the lack of access to data.

Access to real-time channel data provides the transparency and insights needed for advertisers to protect their entire PMax spend. For example, with one of their clients, who had trusted their paid search budget entirely to PMax, they identified where PMax was cannibalizing their branded keywords.

We highlighted that these keywords would perform better in a stand-alone search campaign where you have full control and transparency. Additionally, their client saw PMax bidding on low-performing search terms, rapidly burning daily budgets on terms any experienced search marketer would typically exclude. This was not fraud per-se; it was AI optimizing to suboptimal outcomes. Armed with this data, marketers can then confidently add brand exclusions across PMax (to avoid the wasted overspending on branded keywords) and add account-level negative keywords for extremely poor-performing generic terms.

Suggested newsletters for you

Catch up on the most important stories of the day, curated by their editorial team.

See the best ads of the last week - all in one place.

Our media editor explores the biggest media buys and the trends rocking the sector.

Within PMax, they are seeing a tendency of the algorithm to over-optimize towards paying for clicks from customers you already have – this has been a historical issue with any search PPC campaign but has accelerated with PMax. Existing customers who are “lazy browsers” and use Google to get to the homepage by clicking on a paid ad to get to the login section or to find the customer service details are a honeypot for PMax’s AI bias for clicks and engagement.

It’s not logical for marketers to invest in these outcomes from their existing customers when they could allocate their marketing budgets toward acquiring new customers; it simply adds to the Customer Acquisition Cost (CAC). With one of their clients, they saw over 50% of their invalid traffic via PMax was being funneled into paying for existing customers to come to the website. They have made it possible to allow marketers to dynamically set click frequency thresholds on returning users, as well as filter out bot traffic, forcing the AI algorithm to search for new valid users across PMax.

Regarding outright fraud on PMax, they have identified instances of this occurring. On another of their clients, their data shows 7.5% of PMax clicks were invalid across various campaigns. This included identified fraud or highly suspect user engagement where secondary on-site user behavior was inconsistent with a qualified adult human audience. The good news is that this can be stopped. Real-time analysis via their website tags and a direct Google Ad Manager integration made it straightforward to identify and protect clients spending across the entire PMax campaign. This includes automatically passing to Google PMax the necessary data points required to mitigate invalid traffic so they were no longer targeted. Helping clients acquisition of real audiences and incremental conversions and reducing wasted budgets in the process.

I believe that the industry, as a whole, should have a sense of optimism regarding the potential of what AI can bring to the table in terms of driving better marketing performance. There is value to be realized in automation and operational efficiencies. But, whether it’s PMax

or any other AI-led solutions, marketers must push for algorithmic transparency, invest in independent oversight, and not blindly trust the little black box of algorithms if they truly want to drive the best fraud-free performance. Otherwise, whether they target PMax or any other AI-led advertising solution, bad actors will exploit your trust for their benefit, draining your marketing budgets.

Mathew Ratty is chief executive officer at TrafficGuard.